Many modern banking organizations are highly complex. A “bank” is often a larger structure made up of distinct entities, each subject to different regulatory, supervisory, and reporting requirements. For researchers and policymakers, understanding how these institutions are structured and how they have evolved over time is essential. In this post, we illustrate what a modern financial holding company looks like in practice, document how banks’ organizational structures have changed over time, and explain why these details matter for conducting accurate analyses of the financial system.

Note: As of March 2026, the New York Fed will discontinue the Quarterly Trends for Consolidated U.S. Banking Organizations report. In its place, staff economists will begin producing periodic blog posts that highlight evolutions and developments in the banking sector; this marks the first post in that series.

Inside a Financial Holding Company

To illustrate the structure of a holding company in practical terms, we present a stylized example: the fictional “Central Point Corporation.”  

Central Point Corporation’s Organizational Hierarchy

At the top of the organizational structure is the parent, or top-tier, holding company. This is Central Point Corporation (blue box).

The most common type of holding company is a bank holding company (BHC) which is Central Point Holdings LLC (grey box) in our diagram, which files an FR Y-9LP. These are holding companies that own or control one or more commercial banks or other BHCs. The BHC structure was first established by the Bank Holding Company Act of 1956.

The top-tier holding company, Central Point Corporation, however, is classified as a financial holding company (FHC), and these submit FR Y-9C and FR Y-9LP filings. Financial holding companies were introduced as part of the Gramm-Leach Bliley Act (GLBA) of 1999 as a special type of holding company that can engage in a broader range of financial activities beyond traditional banking, such as insurance and securities underwriting. To qualify as an FHC, a company must derive at least 85 percent of its consolidated gross revenues from financial activities. As you can see from our example, though, an FHC may itself hold a BHC.

Holding companies can have many types of subsidiaries. Some of these are domestic commercial banks, such as national banks and state-chartered member banks (which both file an FFIEC 031). Central Point Corporation owns a national bank (Central Point Bank, N.A.) and a state-chartered member bank (Central Point Bank, New York). Holding companies can also contain nonbank financial institutions (NBFIs, which file an FR Y-11 if domestic, and an FR 2314 if international), firms that engage in financial activities besides banking. For example, CP Capital, Inc. and Central Investments, Inc. are NBFIs.

Other subsidiaries may be foreign. These subsidiaries may be placed in an edge corporation (which files an FR 2886b) that is allowed by its charter to engage in foreign business. In our example, Central Point Corporation owns an edge corporation (Central Point International), which in turn holds a foreign bank (Central Point France (FR)) and a foreign nonbank (Global Market Capital Ltd). This organizational structure has direct implications for how institutions are observed in the data. Different entities within the same holding company are subject to different reporting requirements, file different regulatory forms, and may appear (or not appear) in commonly used datasets. To support research on these institutions, the table below (full version available for download) summarizes several commonly used U.S. regulatory forms and the types of entities required to file them. While not exhaustive, it provides a strong starting point for navigating the regulatory reporting landscape and identifying appropriate sources for given research questions.

Preview: U.S. Regulatory Forms and Corresponding Entities

The Evolution of Financial Institutions

Banking organizations were not always as complex as Central Point, and in fact were previously all organized as BHCs. How did banking organizations change over time? And what were the regulatory developments associated with these changes?

Next, we conduct an analysis to show how the composition of holding company types has evolved over time. In addition to FHCs and BHCs, we will also discuss the development of savings and loan holding companies (SLHCs) and intermediate holding companies (IHCs), which have developed as alternative holding company structures in recent decades. SLHCs own or control one or more savings associations or other SLHCs, while IHCs are established by large foreign banking organizations to hold all of their U.S. non-branch subsidiaries. Throughout our analysis, we will describe how these holding company types differ from one another, and the regulatory developments that have encouraged their growth. 

The chart below shows both the number and share of holding company types (BHCs, FHCs, SLHCs, and IHCs) from 1990 to 2024. 

Evolution of U.S. Holding Company Types

Two patterns stand out. First, the total number of holding companies declined substantially from 6,307 in 1990 to 3,801 in 2024, reflecting consolidation in the banking sector. Despite this decline, BHCs remain the most prevalent holding company type, accounting for roughly 80 percent of all regulated holding companies. 

Second, changes in holding company types have been closely tied to regulatory developments. The introduction of the FHC following the GLBA of 1999 allowed BHCs to expand into a broader range of financial activities. While FHCs represent less than 20 percent of all regulated holding companies, the bottom panel of the chart above shows that they hold a disproportionately large share of total assets. 

SLHCs have remained relatively small in count, falling from 459 in 2012 (when they were first incorporated into the Federal Reserve reporting framework) to 190 in 2024. IHCs are few in number (only 11 as of 2024), but account for a meaningful volume of assets given the size of their parent organizations. 

To understand shifts in the composition of holding company types, the chart below tracks changes in holding companies’ designations over time. 

Entity-Type Switches by Year

The most common switch is from BHC to FHC, as shown in the top panel. 518 conversions occurred immediately upon the passage of the GLBA in 2000. Another 123 firms converted in 2001, after which the pace of conversions slowed considerably. After the global financial crisis (GFC) in 2008, BHC-to-FHC conversions dipped while FHC-to-BHC reversions spiked, as institutions switched their charters. One reason for these conversions may be that Dodd-Frank increased the regulatory requirements and reporting burdens for large FHCs, making it relatively more costly to have this structure.

The middle panel shows switches between BHCs and SLHCs, and the bottom panel shows conversions into IHCs. While the conversions to SLHCs and IHCs occurred after the introduction of the respective holding company types in 2011 and 2016, these switches are less frequent than those between BHCs and FHCs. One driving force behind the shifts between BHCs and SLHCs is the transfer of supervisory authority over SLHCs from the Office of Thrift Supervision (OTS) to the Federal Reserve in 2012, which made such switches easier. The switches from BHCs to IHCs occurred once the IHC structure was first mandated under the 2016 intermediate holding company rule, which requires any FBO with more than $50 billion in total global consolidated assets and at least $50 billion in U.S. non-branch assets to establish an IHC to house its U.S. subsidiaries.

Why This Matters

The organizational structure of financial organizations is critical for researchers and other analysts to understand, as it shapes both what banks do and what we observe in the data. Ignoring this information can lead to mismeasurement, sample selection errors, and misleading conclusions.  

For example, consider a researcher trying to understand differences between banking and non-banking activities of holding companies using FR Y-9C filings. In our stylized example, the top-tier entity, Central Point Corporation, files the FR Y-9C but is designated a financial holding company rather than a bank holding company. If the researcher identifies banking activities as only those related to BHCs, they may inadvertently underestimate the amount of banking activities performed by Central Point Corporation and other economically relevant institutions. Because these firms tend to be the largest and most complex, such misclassifications can systematically bias results.

As a second example, consider an analyst tracking the activities of a single organization over time. Suppose Central Point Corporation initially operated as a BHC and later reclassified itself as an FHC. Such a change would expand the range of activities that the organization is able to conduct, thus altering how it appears in regulatory data. The analyst might observe abrupt changes in reporting variables that reflect reclassification rather than true changes in behavior. Without accounting for organizational transitions, these reporting shifts can be misinterpreted as behavioral effects. We would encourage researchers to use the National Information Center database to confirm the structure of holding companies in their data, which could prevent the misclassifications described here.

Summing Up

In this blog post, we described the structure of modern financial institutions with the help of a stylized example, documented how holding company types and designations have evolved, and explained why this knowledge is important for banking analysis. Understanding these organizational choices clarifies how institutions fit within the broader regulatory framework and is essential for researchers involved in banking analysis.

Lily Gordon is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Lee Seltzer is a financial research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

 
@article{GordonSeltzer2026,
    author={Gordon, Lily and Seltzer, Lee},
    title={Behind the ATM: Exploring the Structure of Bank Holding Companies},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={March 31},
    year={2026},
    url={https://doi.org/10.59576/lse.20260331}
}

You may also be interested in:

U.S. Banks Have Developed a Significant Nonbank Footprint

Reserves and Where to Find Them

Do Payout Restrictions Reduce Bank Risk?

This post presents an update of the economic forecasts generated by the Federal Reserve Bank of New York’s dynamic stochastic general equilibrium (DSGE) model. We describe very briefly our forecast and its change since December 2025. To summarize, growth in 2026 is expected to be more robust, and inflation more persistent, than predicted in December. Stronger investment is the main driver for higher growth, while cost-push shocks, possibly capturing the effects of tariffs, are the key factors behind higher inflation. Projections for the short-run real natural rate of interest (r*) are the same as in December.

Note: The DSGE model forecast is not an official New York Fed forecast, but only an input to the Research staff’s overall forecasting process. For more information about the model and variables discussed here, see our DSGE model Q & A. 

The New York Fed DSGE model forecasts use data released through 2025:Q4, augmented for 2026:Q1 with median forecasts for real GDP growth and core PCE inflation from the March release of the Philadelphia Fed Survey of Professional Forecasters (SPF), as well as the yields on 10-year Treasury securities and Baa-rated corporate bonds based on 2026:Q1 averages up to February 25. Starting in 2021:Q4, the expected federal funds rate (FFR) between one and six quarters into the future is restricted to equal the corresponding median point forecast from the latest available Survey of Market Expectations (SME) in the corresponding quarter. For the current projection, this is the January SME. Note that the DSGE forecasts were produced before the start of the Iran war and therefore do not incorporate its economic impact.

Once again, the economy turned out to be more resilient, and inflation more persistent, than the DSGE model had predicted in December. GDP growth in 2025:Q3 turned out to be about 1.5 percentage points higher than anticipated in the November SPF (which the December DSGE forecast used as a nowcast for 2025:Q3, as the Q3 GDP data were not available at the time due to the government shutdown). Moreover, growth in 2026:Q1, at least according to the current nowcast, is also more than one percentage point higher than the model predicted in December. The model attributes these upside surprises mainly to shocks that drive up investment. These shocks, which in the DSGE lingo are known as MEI (marginal efficiency of investment) shocks, arguably capture the strength of AI-related investment in the second half of 2025 and the beginning of 2026.   

In light of the forecast misses, the model revised upward its projections for growth in 2026 by nearly half a percentage point (1.0 versus 0.6 percent). GDP growth projections are lower than they were in December for the remainder of the forecast horizon, as the level effects of the shocks on output fade (2027, 2028, and 2029 GDP growth forecasts are 0.2, 0.9, and 1.3 percent in March versus 0.8, 1.3, and 1.8 percent, respectively, in the December forecasts). The probability of a recession, defined as four-quarter output growth falling below -1.0 percent over the next four quarters, is 35.8 percent, lower than in December (37.5 percent).   

Core PCE inflation in 2025:Q4 was slightly lower than expected, but the nowcast for 2026:Q1 inflation is almost half a percentage point higher than the model predicted in December. The DSGE attributes this forecast error to cost-push shocks, which possibly capture the effects of tariffs, as well as other idiosyncratic factors affecting inflation. As a consequence of this forecast error, the model revised upward its projections for core PCE inflation (2.4, 1.9, 1.9, and 2.0 percent for 2026, 2027, 2028, and 2029, versus 1.9, 1.6, 1.7, and 1.8 percent, respectively, in December). 

The model’s predictions for the short-run real natural rate of interest (r*) are essentially the same as in December (1.9, 1.6, 1.3, and 1.1 percent for 2026, 2027, 2028, and 2029, versus 2.0, 1.6, 1.3, and 1.2 percent, respectively, in December). Since projections for the path of the nominal FFR are also unchanged relative to December, but inflation projections are higher, the model views the current path of policy as slightly more accommodative than it was in December. 

Forecast Comparison

Forecasts of Output Growth

Forecasts of Inflation

Real Natural Rate of Interest

Marco Del Negro is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Ibrahima Diagne is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Keshav Dogra is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Elena Elbarmi is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Donggyu Lee is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Michael Pham is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{DelNegroDiagneDograElbarmiLeePham2026,
    author={Del Negro, Marco and Diagne, Ibrahima and Dogra, Keshav and Elbarmi, Elena and Lee, Donggyu and Pham, Michael},
    title={The New York Fed DSGE Model Forecast—March 2026},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={March 20},
    year={2026},
    url={https://libertystreeteconomics.newyorkfed.org/2026/03/the-new-york-fed-dsge-model-forecast-march-2026/}
}

You may also be interested in:

The New York Fed DSGE Model Forecast— December 2025

The New York Fed DSGE Model Forecast—September 2025

The New York Fed DSGE Model Forecast—June 2025

In this post we provide a measure of “global” r* using data on short- and long-term yields and inflation for several countries with the approach developed in “Global Trends in Interest Rates” (Del Negro, Giannone, Giannoni, and Tambalotti). After declining significantly from the 1990s to before the COVID-19 pandemic, global r* has risen but remains well below its pre-1990s level. These conclusions are based on an econometric model called “trendy VAR” that extracts common trends across a multitude of variables. Specifically, the common trend in real rates across all the countries in the sample is what we call global r*. The post is based on the discussion of an insightful paper by Lukasz Rachel on the drivers of r* presented at the Brookings Papers on Economic Activity Fall 2025 conference.

Is There a Global R*? Cross-Country Convergence in R*

The chart below plots estimates of r* using macroeconomic data for the eighteen developed countries included in the Jordà-Schularick-Taylor Macrohistory database. It shows that before the 1980s there is a lot of dispersion in r* across countries. But after the late 1980s this dispersion disappears all of a sudden, arguably as a result of financial market integration. Therefore, after the late 1980s, we can actually talk of a global r*, since the trends in real rates are one and the same across advanced countries. The important implication of this finding, which was first documented in “Global Trends in Interest Rates,” is that both the decline in r* from the 1990s to before COVID and the post-COVID rise that is evident from the chart are global phenomena.

Global Convergence in R*

The Decline, and Recent Rise, in the Global (and U.S.) R*

The dashed black line in the chart below shows the posterior median of global r* with the shaded areas showing the 68 and 95 percent posterior coverage intervals. The dotted black line shows the posterior median of U.S. r*. According to the model, global r* fell from about 3 percent in the early 1990s to below 0 percent after the financial crisis. It continued declining in the 2010s and then rose by about 1 percentage point after COVID. By and large, the U.S. r* has tracked global r* since the late 1980s, except that it declined comparatively more in the aftermath of the financial crisis. By 2024, the end of the sample, the median posterior estimates of both global and U.S. r* are around 0.5 percent (more precisely, 0.31 and 0.46). This figure is broadly consistent with “the GDP-weighted average of estimates of r-star for … Canada, the Euro Area, the United Kingdom, and the United States,” according to a recent presentation by New York Fed President and CEO John C. Williams, although the U.S. r* point estimate is a bit below current estimates of U.S. r* from the well-known Laubach-Williams and Holston-Laubach-Williams models, which are about 1 percent. However, the large posterior coverage intervals shown in the chart are there to remind us that extracting trend from cycle is a difficult task, and that one should take point estimates with more than a grain of salt. The 68 percent posterior coverage intervals for both global and U.S. r* range from about -0.5 to above 1 percent, while the 95 percent intervals range from about -1.5 to above 2 percent.

The Global and the U.S. R*

Even if the level of r* is very uncertain, the model is able to detect changes in r* over time with greater statistical confidence. The first row of the table below reports the decline in global and U.S. r* from 1990 to 2019, by our calculations. The median estimate of the decline is about 3.5 percentage points for both the world and the U.S. Although the width of the 95 percent posterior coverage intervals (in parentheses) indicate that the exact magnitude of the decline is uncertain, there is no question statistically that such a decline in r* has taken place from the 1990s to before COVID: the posterior probability that the change is less than zero is greater than 97.5 percent, as indicated by the three stars next to each number.

Both the table and the charts above also point to a statistically significant rise in both global and U.S. r* in the post-COVID period: of about 0.8 percentage point for the global r* and a little more than 1 percentage point for the U.S. r*. It is important to remark that the magnitude of the increase is smaller than that of the pre-COVID decline, hence r* remains well below what it was in the 1990s. To the extent that one believes the model’s message of a post-COVID increase in r*, it begs the question of what is driving it. Since the increase is not just a U.S. phenomenon but global—the first chart in the post shows that r* rose in pretty much all developed countries—its drivers better be global as well. Purely country-specific explanations for the increase in r* may not be the whole story.

In previous research, some of us have argued that an increase in the global convenience yield—that is, the convenience for safety and liquidity that applies to all advanced economies’ government bonds—is an important driver of the pre-COVID decline in r*. In other words, investors’ appetite for safety (and liquidity) drove government bond yields across advanced economies down between 1990 and 2019. To what extent did a sudden decline in the convenience yield between 2019 and 2024 drive r* up?

The bottom panel of the table above decomposes changes in r* into a component attributable to the convenience yield (“cy”) and a component attributable to other drivers (“Other”). The table shows that indeed the increase in the convenience yield explains about one-third of the decline in r* both for the U.S. and the world between 1990 and 2019. The decline in the convenience yield for government bonds also explains one-third to one-half of the post-COVID rise in r*, although it is not precisely estimated. This decline, which in the U.S. is reflected in a compression of corporate bond spreads, reflects the fact that for a variety of reasons, possibly including the surge in government debt across advanced economies, the appeal of government bonds in the U.S. and around the world in terms of safety and liquidity has declined. At the same time, the table shows that this decline is clearly not the entire story: the change in the remainder is larger and statistically more significant than the change in “cy.”

If not the convenience yield, what explains the post-COVID rise in r*?  Two plausible candidates are: a forthcoming artificial-intelligence-driven uptick in productivity growth and future surges in debt-to-GDP, possibly driven by a perceived unwillingness on the part of governments in advanced economies to raise taxes to deal with the demographic transition, and/or by higher expected military spending. Rachel’s Brookings paper considers these scenarios and shows that both factors might well be driving the rise in r*, although the abruptness of the rise is harder to rationalize in the model.

In sum, we find that r* has risen by about 1 percentage point in the U.S. and in advanced economies after COVID, and that about one-third of the change may be due to a decline in the convenience yield for government bonds. The r* estimates discussed in this post, and the replication code, are available on this GitHub page. We hope to update these estimates as new data becomes available.  

Marco Del Negro is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Elena Elbarmi is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Michael Pham is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{DelNegroElbarmiPham2026,
    author={Del Negro, Marco and Elbarmi, Elena and Pham, Michael},
    title={The Post‑Pandemic Global R*},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={February 25},
    year={2026},
    url={https://doi.org/10.59576/lse.20260225}
}

You may also be interested in:

On GitHub: Global R* Estimates

The New York Fed DSGE Model Forecast— December 2025

The Post‑Pandemic r*

Data releases for inflation have been scarce over the past four months due to the government shutdown. As a result, until January 22 no personal consumer expenditures (PCE) data were available beyond September and the consumer price index (CPI) had many missing entries for the one-month changes for October and November. In this post, we use an extended version of the New York Fed’s Multivariate Core Trend (MCT) inflation model to examine changes in underlying inflation over this period. The MCT model is well-suited to do so because it decomposes sectoral inflation rates into a trend (“persistent”) and a transitory component. In contrast to core (ex-food and energy) inflation, its aim is to remove all transitory factors, thus identifying the underlying trend. In addition, since the model can handle missing data—like for October—it can produce values for trend inflation for months where little or no data were released. Our findings suggest caution: while the fragmented data from November initially signaled a deceleration in price pressures, the integration of December data indicates that these reductions were largely transitory. Once the full data set is used, the aggregate trend for December stands at 2.83 percent, an increase from 2.55 percent in September.

Measuring Missing Inflation Data

We start by measuring month-on-month inflation rates for the last half of 2025. We are mainly interested in the rates for October, November, and December, as little to no data came out over this period.

The observation equation of the MCT model links annualized month-on-month sector-specific inflation rates to common and sector-specific trend and transitory components:

where τ_(c,t) and τ_(i,t) are the common and sector-specific trends that follow a random walk, and ε_(c,t) and ε_(i,t) are transitory components that are modeled as moving average processes. The model takes both sectoral CPI and PCE rates as input. The equation above is for sectoral PCE inflation. CPI inflation sectors are modeled similarly, but the same trends and idiosyncratic components as for PCE are included, up to a constant of proportionality that is estimated along with the other model parameters. Using the trend components, we can compute the aggregate trend:

where the weight ω_(i,t) is the expenditure share of sector i. As we are primarily interested in the trend underlying core inflation, the weights for the sectors food, gasoline, and utilities are set to zero. In addition, we compute trend estimates for core goods, nonhousing core services, and housing using a similar aggregation scheme as for the aggregate trend but tailored to the specific categories.

We estimate the model five times based on the increases in data availability described in the following table. The 10/24/25 information set establishes a benchmark by using only the inflation data available during the shutdown. This is used to produce a nowcast for realized inflation in all components for September and corresponding forecasts for October-December. Next, we update the information set and our estimates every time a new CPI or PCE data set is released.

Incorporating the November CPI and PCE releases requires additional care. Specifically, as there was no CPI released for October, and only an estimate for PCE in October, we cannot compute the month-on-month inflation rates for October and November (see the technical note in the January 22 PCE release); only the bimonthly inflation rate can be directly computed from the price indexes for September and November. To ensure that the components are correctly measured, we adjust the observation equation above for this period. Specifically, we adjust the observation equation for the bimonthly inflation measurement in November as follows:

where the unobserved month-on-month inflation rates for October and November, November, π_i,t and π_i,t-1, respectively, are modeled as above. With this modified observation equation, we can back out estimates for the trend and transitory components for October and November just as in the original model. Moreover, we can also back out estimates (that is, nowcasts) for the unobserved monthly PCE inflation rates π_i,t and π_i,t-1 for October and November. Together with a forecast for PCE inflation in December, this helps us answer the question of whether inflation is persistent or temporary.  

The nowcasts and forecasts for headline and core PCE inflation are shown in the following chart, where the different lines describe the estimates for the different information sets. We find that the PCE release for September (12/05/25) had limited influence on the forecasts for October-December when compared to our October baseline. In contrast, the November CPI release (12/18/25) greatly reduces the PCE nowcasts and forecasts for October through December; that is, the November nowcast for headline inflation is nearly 50 percent lower (falling from 0.23 to 0.12) when compared to the benchmark. For core inflation, the drop is slightly lower but still sizable (from 0.23 to 0.13).

November and December Data Gave Opposing Messages About Inflation Dynamics

The December CPI release (01/13/26) changes the optimistic impression that the November CPI provides. Based on December data, the nowcast for November moved upward slightly, but most importantly, the December nowcast now exceeds the benchmark forecasts for December, thus erasing any reductions in inflation suggested by intermediate data releases. This holds both for headline and core PCE inflation.

Is It Trend?

The nowcasts and forecasts that we discussed mix transitory and trend components. This leaves open the possibility that the documented differences stemming from the November and December CPI releases are merely transitory. For instance, measurement error introduced by data collection during or just after the government shutdown may explain the differences. To investigate this possibility, we extract the trend estimates from the MCT for the different data releases. The estimates are shown in the following chart. The top left panel shows the aggregate trends for each release.

November and December Data also Impacted Trend Estimates

We find a similar pattern to that for the PCE nowcasts: the November CPI greatly reduced the trend estimate, whereas the December CPI returns the trend to its benchmark level established by the pre-shutdown information set dated 10/24/25. Not only is the aggregate trend estimate for December roughly unchanged from its October forecast, but it increased relative to the September estimate. Specifically, the December MCT stands at 2.83 percent whereas the September estimate was 2.55 percent.

Looking at the trend components for different broadly defined sectors—core goods, nonhousing core services, and housing—shows that the patterns observed for the aggregate are replicated across sectors. There are differences regarding the post December level, which is either above its September level (core goods and nonhousing services) or below (housing). But the conflicting effects of the November and December CPI releases are visible in each sector-specific trend.

We conclude that the changes documented in the PCE nowcasts and forecasts across the different CPI releases are not solely due to transitory components but are very much driven by changes in the trend components of the model.

Was November CPI Unusual?

Given the findings above, it is natural to ask whether the November CPI and PCE releases were unusual. Indeed, at the time of the CPI release, numerous economists and market analysts raised substantial concerns about the reliability of the release, noting that significant gaps in data collection—caused by the recent federal government shutdown—likely distorted the inflation figures. Many price observations typically gathered in October were missing, and delayed data collection in November meant that some prices were only captured later in the month, including during heavy holiday discounting, which can artificially lower reported inflation. This raised skepticism among experts who warned that the headline slowdown may not reflect true underlying price trends and could be an artifact of the disrupted methodology rather than a genuine shift in inflation dynamics.

To study the impact of the November CPI in more detail, we compare our latest nowcast, based on the information available on 01/22/26, with a nowcast that is computed after setting the November CPI and PCE to missing. The next chart shows the comparison for the PCE nowcasts. We find that PCE headline and core nowcasts omitting the November CPI release imply that headline and core PCE inflation were nearly constant, or slightly increasing over the September-November period.

November CPI/PCE Inflation Was Unusually Low

The large downward revision in the trend estimates leads us to conclude that the November data were indeed unusual, likely reflecting the delays in data collection caused by the shutdown. But because these issues were temporary, we see the trend estimates from our MCT model that incorporate December data as a reliable signal of the current state of inflation. By that measure, inflation was close to 3 percent by the end of 2025.

Martín Almuzara is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Geert Mesters is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{AlmuzaraMesters2026,
    author={Almuzara, Martin and Mesters, Geert},
    title={Seeing Through the Shutdown’s Missing Inflation Data},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={February 17},
    year={2026},
    url={https://doi.org/10.59576/lse.20260217}
}

You may also be interested in:

Multivariate Core Trend Inflation: Estimates and Sectoral Insights

Measurement and Theory of Core Inflation

Over the course of 2025, the average tariff rate on U.S. imports increased from 2.6 to 13 percent. In this blog post, we ask how much of the tariffs were paid by the U.S., using import data through November 2025. We find that nearly 90 percent of the tariffs’ economic burden fell on U.S. firms and consumers.

2025 Tariffs 

In the chart below, we plot U.S. import tariffs by month in 2025. The blue dots depict the average statutory tariff rate, weighted by 2024 annual import values. The red dots show the average duty rate by month, calculated as total duties collected divided by the value of total imports. The average tariff rate was very low at the beginning of the year, at 2.6 percent. It then spiked in April and May, when tariffs on Chinese goods were raised by 125 percentage points, before being reversed by 115 percentage points in mid-May. By the end of the year, the average tariff rate was 13 percent. 

The Average Tariff Rate Has Increased

The average duty rate is lower than the average tariff rate because of the many exemptions granted. For example, although the U.S. levies a 35 percent tariff on Canadian imports, 83 percent of those imports are exempt from U.S. duties under the U.S.-Mexico-Canada Agreement (USMCA). A second reason for the lower average duties is that importers shift away from high-tariffed goods. The difference between the statutory rate and the duty rate peaked in April and May, when importers shifted away from Chinese imports in order to avoid the higher tariffs levied on Chinese goods. 

The next chart shows how global supply chains shifted in response to the higher tariffs. We plot import shares by country (or region) for 2017, 2024, and 2025, and countries are ordered by their 2017 import shares. These seven exporters accounted for approximately 80 percent of U.S. imports in 2017, with Chinese goods making up nearly 25 percent of total imports that year. Following a 9-percentage-point increase in tariffs on Chinese goods levied in 2018 and 2019, Chinese imports fell to around 15 percent by 2024. What is striking is that, in the first eleven months of 2025, China’s share of U.S. imports fell by another 5 percentage points, slipping below 10 percent. In contrast, Mexico and Vietnam gained the most market share. China now faces the highest tariffs among the countries and regions shown in the chart. 

China’s Share of U.S. Imports Has Fallen Markedly

Who Bears the Cost of Tariffs?

Tariff incidence is the technical term for how the costs of a tariff are split between foreign exporters and domestic importers. While importers pay the duty, the “economic burden” of the tariff can be shifted onto exporters if they lower their export prices. We illustrate this effect through a simple example: Suppose foreign exporters charge $100 for a good, and the importing country decides to levy a 25 percent tariff on it. If the foreign price remains unchanged at $100, the duty paid is $25, increasing the import price to $125. In this case, the tariff incidence falls entirely on the importer; in other words, there is 100 percent pass-through from tariffs to import prices, and therefore on U.S. consumers and firms. 

In contrast, the exporter might lower its price in order to avoid losing market share. If foreign exporters respond to the tariff by lowering their price to $80 (i.e., $100 divided by 1.25), the price paid by importers will remain $100 (with $20 in duties paid to the government). In this case, 100 percent of the tariff incidence falls on foreign exporters, who now receive $20 less for the same good; in other words, there is zero pass-through from the tariff since the import price is unchanged. 

Considering an intermediate case, suppose the exporter lowers its price to $96 to absorb some of the cost in response to the 25 percent tariff. The 25 percent tariff is then calculated on the new, lower price, making the tariff-inclusive price the importer pays $120. In this scenario, the lower export price means the exporter pays $4 of the burden, while the higher tariff-inclusive price means the importer pays $20. We define the incidence on the importer as the ratio between the price increase due to the tariff ($120 minus $100) and the total tariff revenues; in this example, the incidence on the importer is 83 percent ($20 divided by $24); the incidence on the exporter (that is, the price decrease they suffer as a ratio of the total revenues from tariffs) is 17 percent ($4 divided by $24). 

Because tariff incidence hinges on how tariffs affect export and import prices, we now focus on estimating the impact of tariffs on these prices. We follow the approach used in our previous study, which analyzed the effect of the 2018-2019 tariffs on prices for goods exported to the U.S. In that earlier work, we regressed the twelve-month percentage change in foreign export prices on the twelve-month percentage change in tariffs. We also controlled for average price changes of finely defined products across all countries, and changes in the average price of imports into any country in any month to isolate the differential effects of the tariff. Our past work found that foreign exporters did not lower their prices at all, so the full incidence of the tariffs was borne by the U.S. That is, there was 100 percent pass-through from tariffs into import prices. 

We now conduct the same analysis for the 2025 tariffs, covering twelve-month changes from January 2024 through November 2025 (the most recent available data). We report the results in the table below. In this analysis, we also allow the pass-through to change for different months in 2025. Our results show that the bulk of the tariff incidence continues to fall on U.S. firms and consumers. These findings are consistent with two other studies that report high pass-through of tariffs to U.S. import prices. 

Tariff Incidence Falls Mostly on U.S. Importers

We highlight two main results. First, 94 percent of the tariff incidence was borne by the U.S. in the first eight months of 2025. This result means that a 10 percent tariff caused only a 0.6 percentage point decline in foreign export prices. Second, the tariff pass-through into import prices has declined in the latter part of the year. That is, a larger share of the tariff incidence was borne by foreign exporters by the end of the year. In November, a 10 percent tariff was associated with a 1.4 percent decline in foreign export prices, suggesting an 86 percent pass-through to U.S. import prices. Given that the average tariff in December was 13 percent (see the first chart), our results imply that U.S. import prices for goods subject to the average tariff increased by 11 percent (13 times 0.86) more than those for goods not subject to tariffs. These higher import prices caused firms to reorganize supply chains, as suggested by the findings presented in the two charts above. 

In sum, U.S. firms and consumers continue to bear the bulk of the economic burden of the high tariffs imposed in 2025. 

Mary Amiti is head of Labor and Product Markets in the Federal Reserve Bank of New York’s Research and Statistics Group.

Chris Flanagan is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Sebastian Heise is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

David E. Weinstein is an economics professor at Columbia University.

 
@article{AmitiFlanaganHeiseWeinstein2026,
    author={Amiti, Mary and Flanagan, Chris and Heise, Sebastian and Weinstein, David E.},
    title={Who Is Paying for the 2025 U.S. Tariffs?},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={February 12},
    year={2026},
    url={https://doi.org/10.59576/lse.20260212}
}

You may also be interested in:

Are Businesses Absorbing the Tariffs or Passing Them On to Their Customers?

Who Pays the Tax on Imports from China?

Why Does the U.S. Always Run a Trade Deficit?

Do Import Tariffs Protect U.S. Firms?

Inflation does not always respond to cost and demand pressures in the same way. When shocks are small, the mapping from costs to prices is roughly proportional—double the shock, double the inflation response. But when the economy is hit by large shocks, this proportionality breaks down. As the recent surge and subsequent decline of global inflation showed, price growth can accelerate—or decelerate—by more than one-for-one relative to the size of the disturbance. Economists refer to this pattern as nonlinear inflation dynamics. In this post, I discuss what these nonlinearities mean, how they relate to the slope of the Phillips curve discussed in a companion post, and how firm-level data can help us understand the mechanisms behind them.

Evidence of Nonlinearities in the Data

The chart below illustrates nonlinear inflation dynamics taken from a recent working paper. It plots a measure of real cost changes for the Belgian manufacturing sector against producer price inflation (PPI) over the period 1999–2023.

The Nonlinear Relation Between Inflation Cost Shocks and Inflation

For small and medium-sized cost shocks, inflation rises roughly one-for-one with costs: the dots align along a straight line. When the economy is under stress, however, this proportionality breaks down. Once cost shocks exceed a threshold—around a 20 percent annual increase in our data—the slope steepens sharply. Double the size of the shock and inflation rises by much more than twofold. That is, inflation becomes more sensitive to shocks when disturbances are sufficiently large. This pattern is exactly what we observe in recent years, when the energy shocks and supply-chain disruptions associated with the COVID-19 pandemic and the war in Ukraine drove exceptionally large increases in producers’ costs.

The State-Dependent Nature of Pricing

Nonlinear inflation dynamics reflect the microeconomics of firms’ pricing decisions—and, in particular, the state-dependent way firms respond to shocks. Simply put, firms do not react uniformly: when shocks are small, many wait; when shocks are large, more firms adjust, and by larger amounts.

To see the intuition, consider a world where firms can adjust prices only at random and with a fixed probability—like in a lottery. In this world, the firms that reprice are not necessarily those facing the largest shocks (those whose prices deviate most from their optimal level). Yet when they do reprice, they update their prices to reflect current economic conditions. These ideas underlie the Calvo model, the workhorse model used in academic and policy analysis. In this framework, the average frequency of price changes does not vary with economic conditions, implying that inflation scales linearly with the size of the shock (a linear Phillips curve). Strict as they may seem, the assumptions behind the Calvo model—particularly the fixed probability of price adjustment—are remarkably consistent with the data in normal times.

Things look different when shocks are large. If adjusting prices is costly—because, for example, it involves updating menus or risks alienating customers—firms reprice only when the benefits outweigh those costs, much as a restaurant rewrites its menu only when prices are far out of date. When pressures are modest, few firms find it worthwhile; when shocks surge, many do. These shifts show up clearly in the data: large jumps in the average frequency of price changes occurred, more recently, during the recent pandemic inflation surge.

Such spikes are inconsistent with the Calvo model, but they are exactly what state-dependent pricing models—such as menu-cost models or models with information frictions—predict. These models capture the idea that when the economy is hit by large shocks, the entire price-setting process speeds up.

Three Facts About Pricing

Using detailed administrative data on thousands of Belgian manufacturing firms, we construct an empirical measure of each firm’s price gap—the percentage difference (in percentage terms) between the price it currently charges and the price it would set if it could adjust freely (its “desired price”). The larger the gap—whether positive or negative—the greater the changes in costs or competitive pressures the firm faces, and the further its current price drifts from the desired one. We thus provide a natural way to study how often and by how much firms adjust in response to shocks of different sizes.

Fact 1: Firms adjust more frequently when their prices drift far from optimal

The chart below shows the relationship between firms’ price gaps and the frequency of price adjustment. The blue line represents the probability density function of price gaps and the red line the fraction of firms that adjust their prices (y-axis) at each point of the price gap distribution. Firms in the tails—those with prices far above or below their desired price level—change prices much more often. The relationship is U-shaped: the further a firm’s price is from its target, the more likely it is to adjust. This is exactly what state-dependent pricing models predict. By contrast, in the Calvo model the probability of adjustment is unrelated to the price gap—its predicted curve would be flat.

The Probability of Price Adjustment Rises with the Size of the Price Gap

Fact 2: Large aggregate shocks shift the entire distribution of price gaps, prompting more frequent price changes

The chart below shows how aggregate shocks reshape firms’ incentives to reprice. Before the pandemic, the distribution of price gaps was centered around zero (blue line). In 2022:Q2—the quarter with the largest jump in firms’ production costs—the entire curve shifts rightward (red dashed line) as producers grappled with surging energy prices and widespread supply chain disruptions, and, as predicted by state-dependent pricing, the share of firms changing prices nearly doubled, leading to a marked acceleration in inflation.

Large Aggregate Shocks Shift the Entire Distribution of Price Gaps, Prompting More Frequent Price Changes

Fact 3: Inflation grows disproportionately larger when large shocks hit

In the chart below, we again group firms by the size of their price gap. Each dot represents the average price gap for a group of firms (x-axis) plotted against their average price change (y-axis). The red dashed line shows a nonlinear fit through this cloud of points.

Inflation Responds Nonlinearly with the Size of the Shocks to Firms’ Desired Prices

At any point along this curve, its slope measures how strongly cost shocks are passed through into prices—the steeper the slope, the stronger the pass-through, and the steeper the Phillips curve. This exercise reveals that the mapping between expected price changes and price gaps is not constant. In fact, we can identify two distinct regions.

The Calvo region. When price gaps are small (roughly +/- 10 percent), pricing behavior looks linear: adjustment frequencies are stable, pass-through is proportional, and the elasticity of price changes with respect to gaps matches the average adjustment frequency. This pattern helps explain why the Calvo model works well in low-inflation environments.

The nonlinear regions. At the tails of the price gap distribution, pricing becomes far more reactive. Firms hit by large shocks exhibit nearly double the usual pass-through due to the significant rise in the frequency of price adjustment. The behavior of these firms mirrors what happens to the broader economy during major disturbances—such as those triggered by the COVID-19 pandemic. During these episodes, the economy is pushed into regions where the slope of the Phillips curve rises, and inflation reacts much more strongly and quickly to cost pressures.

Recognizing when the economy shifts between these regimes is important for decision makers who rely on timely signals about changing inflation conditions. Accounting for these transitions is key to understanding how inflation builds and eventually unwinds.

Simone Lenzu is a financial research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.  

 
@article{Lenzu2026,
    author={Lenzu, Simone},
    title={Does the Phillips Curve Steepen When Costs Surge?},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={February 5},
    year={2026},
    url={https://doi.org/10.59576/lse.20260205}
}

You may also be interested in:

Anatomy (not Autopsy) of the Phillips Curve

The relationship between inflation and real economic activity has long been central to debates in macroeconomics and monetary policy. At the core of this debate is the Phillips curve (PC), which measures how strongly inflation reacts to movements in economic conditions. The steepness of this curve matters enormously for monetary policy: if the PC is steeper, inflation rises faster during booms and falls faster in recessions, which entails central banks having to act more forcefully if they want to stabilize inflation around their target. Prior analysis found astonishingly small estimates of the slope of the PC, which suggests that the curve is “flat” (or even dead). In this post, I present evidence from coauthored research showing that, contrary to the conventional view, the Phillips curve is alive and steep, and it captures inflation volatility remarkably well once real marginal cost is used instead of standard real economic activity measures.

The Conventional Formulation of the Phillips Curve

The Phillips curve links inflation to expectations of future inflation and a measure of economic slack. In the conventional formulation of the PC, economic slack is typically proxied by the output or unemployment gaps—the deviation of output or employment from its natural level:

In this view, inflation rises when the economy overheats and falls during slowdowns. The slope of the Phillips curve, K in the equation above, captures how sensitive inflation is to these fluctuations. A large body of research finds that the PC is quite flat—the slope is very small—implying that inflation hardly moves in response to shifts in output or employment gaps. These findings have long puzzled economists and fueled debate about how active monetary policy should be to steer inflation.

The Primitive (Cost-Based) Formulation of the Phillips Curve

There is another way to think about inflation dynamics and its drivers. At its foundation, the Phillips curve emerges from the aggregation of firms’ pricing decisions in response to changes in production costs. In the primitive formulation of the PC, the variable influencing inflation is real marginal cost in percentage deviation from trend, rather than the output or unemployment gap:

In this view, inflation rises when economic forces increase firms’ production costs. The slope of the curve reflects how strongly (and quickly) these costs are passed through into output prices.

The Slope of the Cost-Based Phillips Curve

The cost-based formulation makes it easier to understand the key forces that determine the slope of the PC. In theory, if markets were perfectly competitive and frictionless, prices would move one-for-one with costs: any increase in wages or input prices would be instantly reflected in consumer prices. That is, the slope would be one.

Reality is very different. Firms typically adjust prices infrequently, since doing so involves both direct costs (such as relabeling or updating systems) and indirect costs (such as confusing customers or losing goodwill). In addition, firms often set prices strategically, choosing to delay changes until they are sure cost pressures will last, or timing revisions to match competitors. Thus, frictions that lead to infrequent price changes and strategic considerations in price setting weaken the transmission of cost shocks into prices. The more firms deviate from the ideal of flexible, competitive markets, the flatter the Phillips curve becomes.

Estimating the Slope of the Cost-Based Phillips Curve Using Microdata

How steep is the cost-based PC in the data? Answering this question is notoriously difficult, particularly when the estimation is solely based on aggregate data in which many shocks influence inflation and real activity at the same time. To address this problem, we turn to firm-level evidence. In a recent paper coauthored with Mark Gertler (New York University), Luca Gagliardone (Yale University), and Joris Tielens (National Bank of Belgium), we use detailed microdata on prices and costs to study how individual firms adjust their prices in response to changes in production costs. This approach allows us to quantify how nominal rigidities (infrequent price changes) and real rigidities (strategic interactions among firms) dampen the response of prices to cost shocks. With these estimates we recover the slope of the primitive form of the Phillips curve.

Our analysis suggests a strong link between inflation and real economic conditions, as captured by producers’ costs. On average, firms keep prices fixed for three to four quarters, confirming a substantial degree of nominal rigidity. We also find strong evidence of strategic complementarities: firms adjust less aggressively because they prefer to move in step with competitors, which cuts the pass-through of cost shocks roughly in half. Taking these frictions into account, we estimate the slope of the cost-based Phillips curve to be three to ten times larger than the estimates for the output- or unemployment-based PC formulation. This implies that the Phillips curve is steep, not flat, even in normal times.

Accounting for Aggregate Inflation Dynamics

Cost pass-through plays a dominant role in shaping aggregate inflation. To illustrate this, we build a cost index by combining firm-level changes in labor and input costs and then feed it into the cost-based Phillips curve. Using the Belgian manufacturing sector as a case study, we construct a model-generated inflation series by feeding data on costs into the Phillips curve. The results, shown in the chart below, show that the predicted inflation aligns very closely with actual producer price inflation (PPI) in Belgium. The two series are highly correlated, with a correlation coefficient above 80 percent; quantitatively, movements in production costs alone account for about 70 percent of observed inflation fluctuations, highlighting the central role of costs in driving inflation dynamics.

Reconciling the Steep Cost-Based Phillips Curve with the Flat Output-Based Phillips Curve

These results raise a natural question: Why does the cost-based Phillips curve slope steeply while the output-based one appears flat? And are these findings at odds with previous research?

Our research shows that the “flatness” of the conventional Phillips curve reflects a weak link between output gaps and marginal costs. In other words, while production costs feed directly into firms’ pricing decisions, movements in output (or unemployment) bear only a loose relationship to inflation.

From a conceptual standpoint, the conventional PC holds only if marginal cost and the output gap move proportionally—an assumption that requires, among other things, perfectly flexible wages. When these conditions fail, the output gap may be a poor proxy for real marginal cost, biasing estimates of PC downward. Moreover, even if proportionality roughly holds, the output-based slope equals the cost-based slope scaled by the elasticity of marginal cost with respect to output. If this elasticity is low, the slope of output-based PC will be low as well, even if the slope of the cost-based PC is sizable.

These results are confirmed in the data. Focusing on the pre-pandemic period (1999–2019), we estimate a very low elasticity of marginal cost with respect to output. This finding helps explain why the cost-based PC is steep while the conventional PC looks flat.

Lessons for the Post-Pandemic Inflation Surge

Our findings show a strong pass-through from marginal costs to prices, which explains why the cost-based Phillips curve matches inflation dynamics so well. The weak link between output and marginal cost, on the other hand, helps explain why the conventional output-gap version of the PC looks “flat.” In normal times, two factors drive this low elasticity: firms’ cost schedules tend to display nearly constant short-run returns to scale, so marginal costs barely move with output; and wage rigidity further dampens any feedback from demand to costs.

The pandemic and its aftermath revealed how quickly these relationships can change under stress. Severe shocks—whether from labor market tightness or supply chain bottlenecks—pushed firms against capacity limits, sending marginal costs sharply higher and fueling inflation. At the same time, the slope of the Phillips curve itself can shift. Pre-pandemic data showed stable adjustment frequencies and an approximate linear relationship between inflation and (percentage) changes in real marginal costs. More recently, however, firms have been adjusting prices much more often, raising the elasticity of inflation with respect to costs and generating nonlinear inflation dynamics. I will talk about nonlinear inflation dynamics—what it means, how it works, and what it implies—in a companion post on Liberty Street Economics.

Simone Lenzu is a financial research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.  

 
@article{Lenzu2026,
    author={Lenzu, Simone},
    title={Anatomy (not Autopsy) of the Phillips Curve},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={February 4},
    year={2026},
    url={https://doi.org/10.59576/lse.20260204}
}

You may also be interested in:

Does the Phillips Curve Steepen When Costs Surge?

Global Trends in U.S. Inflation Dynamics 

It is common for price measures to consider changes in quality. That is, a price index might fall even though listed prices are unchanged because the quality of the item has improved. An adjustment for quality captures the fact that consumers are effectively getting more for the same dollar when product quality rises. In practice, however, it is notoriously difficult to measure quality changes since it requires access to detailed data on all product characteristics that matter to consumers. We offer a novel method to infer quality changes and apply it to U.S. import price indices. When we account for quality improvements in this way, we find that the import price inflation based on official measures has been overstated, revealing that consumers have been getting more from their purchases of imported goods than what standard quality adjustments suggest.

Adjusting for Quality

So how can one address not having detailed information on product characteristics? In a recent paper we estimate quality change for a product by looking at changes in consumer demand for imports that are not accounted for by changes in prices.

Take a simple example. Our approach starts by estimating the elasticity of substitution between imports and domestically produced goods. That is, how responsive is the demand for imported goods if those prices rise by 10 percent relative to the producer price index (PPI) for domestic alternatives. An estimated elasticity of 1 would mean that the quantity sold would fall by 10 percent. If the data show that quantities sold only fell by 5 percent after that kind of jump in relative prices, then one can infer there was a partially offsetting improvement in quality. With that information, we can calculate an alternative import price index. Note that the PPI data are adjusted for quality improvements for several industries and we use this as a quality benchmark for adjusting import prices.

The next stage is to calculate import price indices for 155 individual industries, with the estimated elasticity of substitution allowed to vary across industries. These price series are then aggregated to create an overall import price index, seen in the chart below. While the published import price index rose 13 percent from 1989 to 2018, our alternative index fell 8 percent.

U.S. Import Prices Trend Lower When Adjusted for Higher Quality

One can further decompose quality improvements by industry, to examine which industries have been the most important drivers of higher quality. We find, not surprisingly, that imported machinery and electronic equipment (which includes computers and peripheral equipment) exhibit the strongest quality improvements.

Much of the Understated Quality Improvement Is in the Machinery and Electronics Industry

The chart below shows the ratio of import to domestic prices rises without our quality adjustment and falls with our adjustments. This result resolves a puzzle—U.S. data show import volumes of these products moving higher even though import prices have been trending higher relative to domestic prices in the published data. With our quality adjustment, the costs of imported machinery and electronics are no longer rising relative to domestic prices, helping to explain the ever-increasing demand for these imported goods.

Quality Adjustment Implies Falling Relative Prices of Imported Computers Relative to Domestically Produced Alternatives

Conclusion

This post deals with the issue of how to adjust price data for changes in quality. The approach presented here addresses this challenge with a novel method that uses customs records to infer quality change. The analysis shows that import price indices appear to substantially overstate import price inflation by failing to fully capture quality improvements.

Danial Lashkari is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group. 

 
@article{Lashkari2026,
    author={Lashkari, Danial},
    title={Understating Rising Quality Means Import Price Inflation Is Overstated},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={January 14},
    year={2026},
    url={ https://doi.org/10.59576/lse.20260114}
}

You may also be interested in:

Measuring Price Inflation and Growth in Economic Well‑Being with Income‑Dependent Preferences

The R&D Puzzle in U.S. Manufacturing Productivity Growth

Among people in prime working age (25-54), around 7 percent have a disability of some kind. In this set of companion posts, we will examine how prime-aged workers with disabilities have fared in the labor market compared to the year prior to the pandemic. In this first post, we will show that people with disabilities are far less likely to be employed than people without disabilities, with both lower labor force participation and higher unemployment playing a role. We will also show that although employment rates of people with disabilities are very low, they have risen rapidly during the post-pandemic period, largely because of rising labor force participation. Our results are consistent with an increased prevalence of work from home (WFH) arrangements in the post-COVID period differentially benefiting people with disabilities.

We obtain data on employment and disability status from the Current Population Survey (CPS). Respondents are asked to indicate whether or not they have a disability, and if they do, whether it is a vision, hearing, memory, physical, mobility, or care disability. It is important to note that respondents in the CPS are a representative sample of the civilian noninstitutionalized population and therefore people with disabilities are represented in proportion to their prevalence in the population aged 25-54 years old.  

The chart above presents employment rates for people without disabilities aged 25-54 and for the people with disabilities in that age group. We see that while more than 83 percent of the population aged 25-54 who do not have a disability were employed in August 2025, only about 45 percent of people with disabilities aged 25-54 were employed. The employment gap between people with and without disabilities is far larger than any of the racial, ethnic, and gender gaps, and is even larger than the employment gap between people with a bachelor’s degree and people who did not graduate from high school. 

However, the employment rate for people with disabilities was lower before the pandemic. In January 2019 it stood at a little under 37 percent, more than eight percentage points below its current level. During the pandemic, the fraction of people with disabilities who were employed tumbled to a low of 32.6 percent but then began a secular rise to exceed 47 percent in July 2024. Since that point, the employment rate for people with disabilities has decreased slightly to its present level but still remains far above its pre-pandemic numbers. 

Using the CPS, we can look at the employment rates for people with specific disabilities. As the above chart shows, as of August 2025, people with hearing and vision disabilities had employment rates above 50 percent, while people with memory disabilities had an employment rate of 43 percent. People with physical disabilities were employed at a rate of 29 percent, people with mobility or care difficulties at 22 percent (we combine these two groups to obtain more precise estimates and as these disabilities are substantially similar). Employment rates increased after the pandemic for nearly every disability category, although the rise was more pronounced for people with memory and physical difficulties. This increase may have been contributed to by two factors: the increased incidence of work from home and long COVID.  

First, expanded opportunities to work from home that arose during the pandemic and beyond may have been especially salient for workers with physical and care disabilities, opening up remote work opportunities and consequently resulting in increases in their employment rates. This factor played a positive role in employment rates for workers with other disabilities as well. A second reason may be that more prime-aged workers now have disabilities due to long COVID—for example, brain fog could be a “memory disability”—but remain employable, more so in a remote work setting, driving the employment rate of people with disabilities upward. In fact, the fraction of prime-aged with memory disabilities increased from 2.5 percent to 3.5 percent of the prime-age population after the beginning of the pandemic. The decline of employment rates for people with disabilities starting mid-2024 is consistent with the gradual curtailment of WFH opportunities in the workforce, as the fraction of workers engaged in telework fell from a peak of 23.8 percent in October 2024 to 22.1 percent in August 2025. 

Is the convergence of the employment gap between workers with and without disabilities that followed the pandemic driven by a convergence of labor force participation (LFP) rates or of unemployment rates? In the bottom panel of the above chart, we see a visible convergence of the labor force participation gap until mid-2024 (with the difference between the LFP of workers with and without disabilities declining by 10 percentage points), with the decline driven by an increase in the labor force participation of people with disabilities.  

In the top panel, we see a relatively constant gap in unemployment rates of workers with and without disabilities of roughly five percentage points since mid-2022, with few fluctuations over time. Thus, the convergence in employment rates between people with and without disabilities is driven by the convergence in their labor force participation rates. The labor force participation rate of people with disabilities peaked in July 2024 and subsequently declined by two percentage points, driving most of the concurrent decline in their employment rate. Nevertheless, unemployment is also a serious problem for people with disabilities, who experience unemployment rates of 8 percent, more than twice the rates for people without disabilities. So differences in both LFP and unemployment rates contribute to the employment gap faced by workers with disabilities. 

People with disabilities have some of the worst labor market outcomes among the groups we have considered in the Economic Heterogeneity Indicators (EHIs), but this picture has been evolving. The significant increase in employment for people with disabilities during the post-pandemic period may be a consequence of the interplay of two factors: increased remote-work opportunities and an increase in the fraction of people with disabilities due to COVID. However, it also suggests that the low labor market participation of people with disabilities may not be a constant and we will continue to monitor their labor market situation in future releases of the EHIs. In our companion post, we will discuss the labor market earnings of people with disabilities. 

Rajashri Chakrabarti is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.  

Thu Pham is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Beck Pierce is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Maxim Pinkovskiy is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{ChakrabartiPhamPiercePinkovskiy2026,
    author={Chakrabarti, Rajashri and Pham, Thu and Pierce, Beck and Pinkovskiy, Maxim L.},
    title={Disability in the Labor Market: Employment and Participation},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={January 12},
    year={2026},
    url={https://doi.org/10.59576/lse20260112a}
}

You may also be interested in:

Disability in the Labor Market: Earnings

Long COVID Appears to Have Led to a Surge of the Disabled in the Workplace

Do Veterans Face Disparities in the Labor Market—And What Accounts for Them?

Why do some entrepreneurs drive economic growth while others do not? This piece discusses new work that studies entrepreneurs using a comprehensive dataset from Denmark. We study who becomes an entrepreneur, along with their hiring and business decisions, and find that a distinct minority are “transformative.” These individuals, who generate disproportionate productivity gains, tend to have high IQ scores, be well-educated, and hire technical (R&D) workers. The data support the idea of productivity growth being driven by the symbiotic relationship between transformative entrepreneurs and R&D workers. For policymakers, the lesson is that when an economy has more R&D workers and transformative entrepreneurs, they sustain higher long-run productivity growth.

Facts on Entrepreneurs 

Our work develops new facts using comprehensive data on individuals and firms from Denmark Statistics. For instance, individuals’ IQ test scores come from a military test that is generally taken in very early adulthood. We combine this measure with details of entrepreneurs’ education and parental backgrounds and firm performance to connect individuals’ backgrounds with their firms. The data are discussed in further detail in our working paper. 

We identify entrepreneurs as individuals who register as primary founders of firms with at least one employee. Among this group, we distinguish transformative entrepreneurs, those who hire at least one R&D worker, from non-transformative entrepreneurs, who do not employ technical personnel. This classification captures innovation intent rather than performance outcomes. R&D workers are defined as individuals employed in occupations with high patenting intensity, reflecting their direct involvement in innovation-related activities. 

The data reveal striking patterns. The first chart shows how the shares of entrepreneurs and R&D workers vary across IQ deciles. The pattern is reversed across the two groups: individuals with higher IQs are more likely to be R&D workers, but less likely to be entrepreneurs. 

R&D Workers Tend to Have Higher IQs than Entrepreneurs

The pattern flips for transformative entrepreneurs, as shown in the next chart: individuals with higher IQs are more likely to be transformative entrepreneurs. These patterns remain when controlling for parental background, education, and age. 

Transformative Entrepreneurs Have Higher IQs than Other Entrepreneurs

Educational attainment shows similar patterns as IQ: college graduates are less likely to become entrepreneurs overall but are much more likely to found transformative firms once they do. The strongest predictor of entrepreneurship is having a parent who is an entrepreneur. Education is central for transformative entrepreneurship and R&D workers, and is a key building block for an innovative ecosystem. 

The Firms of Entrepreneurs 

Tracking firm employment over the first decade after a firm is founded reveals substantial performance differences between entrepreneur types. Transformative entrepreneurs start firms that employ approximately twice as many workers as those founded by non-transformative entrepreneurs and maintain significantly higher annual employment growth rates (16 percent versus 8 percent). This can be seen in the next chart.  

Transformative Firms Grow Faster

Revenue trajectories mirror employment patterns. Transformative firms reach seven times their initial revenue by year ten, while non-transformative firms plateau at 2.5 times initial levels. Exit rates differ modestly between firm types: in their first year, 16 percent of non-transformative firms exit, compared to 12 percent of transformative firms, though both rates converge to around 8–9 percent by year eight. 

Industry sorting by IQ also exhibits intuitive patterns. Low-IQ entrepreneurs concentrate in traditional sectors like trade/transport and construction/agriculture, while high-IQ entrepreneurs dominate knowledge-intensive industries, with the knowledge/communication sector representing 50 percent of entrepreneurial activity in the top IQ deciles.

The Macroeconomics of Entrepreneurship 

We develop a quantitative model to study the macroeconomic effects of entrepreneurship on innovation and economic growth, incorporating individual differences in ability, family income, and entrepreneurial exposure. In the model, individuals choose between production work and R&D careers initially, then decide later in life whether to become an entrepreneur, consistent with the age profile in the data. In line with the data, educational attainment, necessary for a technical career path, is determined by both ability and family resources. The model replicates key empirical patterns: negative selection into general entrepreneurship with respect to ability, positive selection into transformative entrepreneurship and R&D work with respect to ability, and the observed correlation between firm performance and entrepreneur type. 

The model captures the symbiotic relationship between transformative entrepreneurs and R&D workers by linking individual career choices to firm-level innovation decisions. This microeconomic foundation enables us to trace how barriers and choices at the individual level aggregate to economy-wide productivity effects through both supply-side constraints (shaped by the availability of R&D workers) and demand-side limitations (determined by the pool of transformative entrepreneurs hiring technical talent). This framework is necessary to develop a full range of realistic responses to policies and changes in the institutional environment. 

We use the model to study the effects of various constraints and policies. Eliminating financial constraints in education access, for example, increases R&D workers’ share of the workforce by 15 percent and boosts transformative entrepreneurs’ share by 7 percent, with both effects concentrated among high-ability individuals from low-income families. In addition, the pace of economic growth increases by 11 percent. 

Supply-side changes alone generate only 60 percent of the total growth effect, with the remaining 40 percent stemming from increased demand for R&D workers by additional transformative entrepreneurs. This illustrates the symbiotic relationship between these entrepreneurs and R&D workers.  

Policy-wise, subsidies to technical education outperform R&D and startup subsidies, generating around four times as much initial return at low budgets by unlocking access to innovative careers for talented individuals from disadvantaged backgrounds. These simulations suggest that it is sensible to mix the different subsidies to jointly stimulate the supply and demand sides of the innovation pipeline. 

Conclusions 

This piece explores the macroeconomics of entrepreneurship using rich microdata. The evidence illustrates that differences between entrepreneur types are stark, and this has substantial implications for aggregate innovation and growth. 

The symbiotic relationship between transformative entrepreneurs and technical workers means that policies affecting one group also influence the other. Education subsidies prove most effective because they simultaneously increase both the supply of R&D talent and demand for R&D talent from the pool of innovation-focused entrepreneurs. Policies that develop ecosystems of talent foster connections between the builders of companies and the builders of ideas. This lesson applies broadly to innovation ecosystems: a thriving economy emerges from many interconnected elements working together. 

Ufuk Akcigit is a professor of economics at the University of Chicago.

Harun Alp is a principal economist in the Emerging Market Economies section of the Board of Governors of the Federal Reserve System.

Jeremy Pearce is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

Marta Prato is an assistant professor of economics at Bocconi University.

 
@article{AkcigitAlpPearcePrato2026,
    author={Akcigit, Ufuk and Alp, Harun and Pearce, Jeremy and Prato, Marta},
    title={Which Entrepreneurs Boost Productivity?  },
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={January 5},
    year={2026},
    url={https://doi.org/10.59576/lse.20260105}
}

You may also be interested in:

How Firms Spread Good Management

Do Employers Comply with Pay Transparency Requirements in Job Postings?

U.S. Banks Have Developed a Significant Nonbank Footprint

This post presents an update of the economic forecasts generated by the Federal Reserve Bank of New York’s dynamic stochastic general equilibrium (DSGE) model. We describe very briefly our forecast and its change since September 2025. To summarize, growth in 2025 is expected to be stronger than in September due to a lower projected path of the policy rate, as well as higher productivity. Inflation projections are higher in 2025 because of cost-push shocks, which capture the effects of tariffs. The model’s predictions for the short-run real natural rate of interest (or r*) in 2025 have decreased relative to September.

Note: The DSGE model forecast is not an official New York Fed forecast, but only an input to the Research staff’s overall forecasting process. For more information about the model and variables discussed here, see our DSGE model Q & A. 

The New York Fed DSGE model forecasts use data released through 2025:Q2, augmented for 2025:Q3 and 2025:Q4 with the median forecasts for real GDP growth, core PCE inflation, and short-run inflation expectations from the November release of the Philadelphia Fed Survey of Professional Forecasters (SPF) for 2025:Q3 and 2025:Q4, as well as the yields on 10-year Treasury securities and Baa-rated corporate bonds based on 2025:Q4 averages up to November 25. Starting in 2021:Q4, the expected federal funds rate (FFR) between one and six quarters into the future is restricted to equal the corresponding median point forecast from the latest available Survey of Market Expectations (SME) in the corresponding quarter. For the current projection, this is the October SME. 

Growth in 2025 is forecasted to be almost half a percentage point stronger than in September (1.8 versus 1.4 percent). This change in the forecast is due to yet another upside surprise in economic activity: just as activity in 2025:Q2 was stronger than had been projected in March, GDP growth in 2025:Q3, at least according to the November SPF, turned out to be higher than anticipated in the August SPF (which the September DSGE forecast used as a nowcast for 2025:Q3). The model attributes this upside surprise to a more accommodative projected path of the policy rate than forecasted in September (this change reflects the adjustment of SME policy expectations between July and October), and to an unexpected increase in the level of productivity. GDP projections are slightly lower for 2026, as the level effect of the shocks on output fades, and unchanged for 2027 and 2028 (2026, 2027, and 2028 GDP growth forecasts are 0.6, 0.8, and 1.3 percent in December versus 0.9, 0.8, and 1.3 percent, respectively, in the September forecasts). The probability of a recession, defined as four-quarter output growth falling below -1.0 percent over the next four quarters, is 37.5 percent, roughly the same as in September (33 percent).  

Core PCE inflation in 2025:Q4, at least according to the SPF nowcast, was higher than predicted by the model in September. The DSGE attributes this forecast error to cost-push shocks, which arguably capture the effects of tariffs. As a consequence of these shocks, core PCE inflation projections are slightly higher than they were in September for 2025 and 2026 (3.0 versus 2.8 percent for 2025, and 1.9 versus 1.8 percent for 2026). However, inflation projections are a bit lower for the reminder of the forecast horizon (1.6 versus 1.7 percent for 2027, and 1.7 versus 1.8 percent for 2028). 

The model’s predictions for the short-run real natural rate of interest (r*) have decreased somewhat in 2025 (2.2 versus 2.6 percent) because of the level increase in total factor productivity (TFP), which the model views as largely temporary, but are essentially unchanged for the remainder of the forecast horizon (2.0, 1.6, and 1.3 percent, respectively, in 2026, 2027, and 2028 versus 2.1, 1.6, and 1.4 percent in the September forecast). As projections for r* have declined along with projections for the policy rate, the model’s assessment of the stance of policy has not changed much since September. 

Forecast Comparison

Forecasts of Output Growth

Forecasts of Inflation

Real Natural Rate of Interest

Marco Del Negro is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Ibrahima Diagne is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.  

Keshav Dogra is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Elena Elbarmi is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Donggyu Lee is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Michael Pham is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{DelNegroDiagneDograElbarmiLeePham2025,
    author={Del Negro, Marco and Diagne, Ibrahima and Dogra, Keshav and Elbarmi, Elena and Lee, Donggyu and Pham, Michael},
    title={The New York Fed DSGE Model Forecast— December 2025},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={December 12},
    year={2025},
    url={https://libertystreeteconomics.newyorkfed.org/2025/12/the-new-york-fed-dsge-model-forecast-december-2025/}
}

You may also be interested in:

The New York Fed DSGE Model Forecast—September 2025

The New York Fed DSGE Model Forecast—June 2025

The New York Fed DSGE Model Forecast—March 2025

Following the recent passage of legislation in the U.S., payment stablecoins seem to be on the brink of wider-scale adoption and explosive growth in market capitalization. In this post, we contend that the driving factor is not their proximity to digital cash instruments, but rather how they are transferred—via global, open-access, peer-to-peer systems, or “permissionless blockchains,” for short.

The Current State of Stablecoin Activity 

Stablecoin market capitalization has recently exceeded $260 billion and total transfer value reached $27.6 trillion in 2024, surpassing the combined value processed by Visa and Mastercard. On the surface, this appears to represent a fundamental shift in how value is transferred digitally.  

However, research by Visa and Allium Labs suggests that less than 10 percent of stablecoin transaction volumes are organic (that is, come from real people). Instead, they find that most stablecoin activity is linked to bot-like transactions, a reflection of factors inherent to blockchain-based systems. Automated market makers, decentralized exchanges, and algorithmic trading strategies generate significant transaction volume without representing actual payment flows. In addition, non-economic transfers could be made for purposes of obfuscation, privacy, or even the manipulation of on-chain statistics. Finally, the composable and transparent nature of decentralized finance protocols means that a single payment can trigger multiple, observable on-chain movements. These facts highlight the critical distinction between transaction value and actual payment adoption. While blockchain infrastructure is processing massive amounts of value, use cases for consumer and business payments are still developing. 

Notwithstanding these caveats, the growth in legitimate on-chain transactions is substantial. Excluding bot-like transactions, the annual volume of stablecoin transactions rose from $3.29 trillion in 2021 to $5.68 trillion in 2024, a roughly 80 percent increase. This growth demonstrates the practical utility of stablecoins for real-world payments.  

What Does the Future Hold?  

The landscape of real-time payments has become increasingly competitive with the launch of new, faster payment systems—including FedNow, the Federal Reserve’s fast retail payment system. FedNow enables near-instant money transfers, in contrast to the hours or even days traditional U.S. payment systems sometimes require. In addition, it’s cheap, runs 24/7, and connects customers across thousands of financial institutions. 

FedNow’s technical capabilities align closely with many of the benefits commonly associated with stablecoins. On FedNow, funds are transferred directly from the sender’s bank account to the receiver’s bank account, providing immediate finality and reducing counterparty risk for retail customers. This architecture addresses many of the speed and settlement issues that have historically plagued traditional payment systems, and challenges the value proposition of stablecoins as domestic payment solutions.  

Other faster payment solutions that exist today include Real-Time Payments (a fast payments platform supporting Zelle), same-day ACH, and peer-to-peer payment platforms (such as Venmo and CashApp). These systems vary in settlement type, payment rail, costs, payment protocols, and transaction value limits. Each serves different market segments and use cases. 

So, what differentiates stablecoins from these alternatives? The answer lies in accessibility. All the faster payment solutions, including FedNow, operate within the traditional banking system, requiring both sender and receiver to be customers of participating financial institutions. Unbanked and underbanked populations are excluded, and international transactions for which correspondent banking relationships are required are typically not supported. There are initiatives like the BIS Innovation Hub’s Project Nexus that seek to connect faster payment systems around the globe. However, these efforts are not yet fully operational. In contrast, stablecoins are already accessible worldwide and enable transactions without restrictions on value or counterparties. The borderless nature of stablecoins provides significant advantages for international payments, remittances, and cross-border commerce.  

These advantages are not intrinsic to the form of money that stablecoins embody but instead arise from their issuance on permissionless blockchains. Indeed, any monetary instrument issued on such blockchains can achieve borderless, global reach. New types of tokenized money instruments are emerging daily, spanning non-fiat-backed stablecoins,  tokenized Treasury funds, and deposit tokens, each offering benefits that will attract different segments of the market.  

Permissionless Blockchains as Public Infrastructure  

What attributes should permissionless systems possess to support the exchange of tokenized money and assets? We identify three key criteria: 

1. Universal access to settlement. Allowing for payment instruments to be transferred across digital wallets self-managed by individuals provides opportunities for all who are seeking dollar settlement. This takes full advantage of the global, open-access community of blockchain systems.  

2. Code without constraint. Smart contracts can automate complex payment flows, enable conditional payments, and integrate with decentralized applications. This programmability opens new possibilities for business process automation and financial innovation. While programmability is not unique to permissionless blockchains, let alone to distributed ledger technology, permissionless blockchains stand out as anyone can program, and compose a program with any other asset on the blockchain.  

3. Innate composability. Composability means assets and contracts can freely reference and interact with each other. While it is possible to compose across multiple asset classes in traditional finance, this requires significant coordination and consensus across stakeholders. In contrast, on permissionless blockchains composability arises as an innate and defining feature.  

There remain substantive issues that must be resolved before permissionless blockchains can reach mainstream adoption.  

• Permissionless blockchains, unlike instant payment systems, are not regulated by central banks and are not fully integrated into established financial ecosystems. This instills a level of insecurity and mistrust that many tokenized solutions experience due to their decentralized nature.  

• Making payments on permissionless blockchains requires individual users to understand nuanced aspects of blockchain settlement and the novel risks associated with the custody of on-chain assets. Major efforts to streamline the user experience are underway by creating standards that improve account management (for example, with EIP-7702) and incorporating authentication processes that are already familiar to the general public (for example, through implementations of zkLogin).  These initiatives offer more flexible and natural ways for users to secure and access their assets on-chain, and ultimately lower the burden of account management. 

• Consumers require transactional privacy. Since public blockchains provide a transparent record of all transactions, transactional privacy requires some form of end-to-end encryption. Several blockchains offer transactional privacy using cryptographic techniques. However, institutional users often require higher standards of privacy and control, requirements that certain platforms aim to achieve. 

In addition to these usability issues, there are safety, resiliency, and efficiency issues that permissionless blockchains must resolve. 

• They must prevent illicit financial activities, such as money laundering and terrorism financing. These require both knowing and tracking the identities of users (e.g. know-your-customer, or KYC, capabilities), as well as transaction monitoring (e.g. anti-money-laundering, or AML, processes). Implementing these capabilities remains challenging for permissionless blockchains. Identity verification conflicts with permissionless systems; censoring select transactions can be difficult due to their decentralized nature. 

• They must meet high operational standards, and for those that become systemically important, undergo supervision. This poses challenges for distributed networks where no central entity can bear responsibility or represent their stakeholders. 

• To fully integrate with and meet the demands of traditional finance systems, public blockchains require significantly higher throughput capabilities, on par with or exceeding the thousands of transactions per second processed by established systems. Various scalability solutions, including modifications to Layer 1 blockchains (e.g., sharding, proof-of-stake) and Layer 2 solutions (e.g., rollups, state channels), are being developed to bridge this gap. However, the simultaneous achievement of optimal decentralization, security, and scalability remains a significant challenge. 

Final Words 

Recent legislation provides regulatory clarity on payment stablecoins, but more importantly reflects a rising interest in allowing people to exercise greater control over their money. This control was largely surrendered with the ascendance of book-entry money, where payments relied on the updating of proprietary data sets. In some respects, permissionless blockchains represent a return to peer-to-peer transfers of value, albeit in digital form. We have argued that this can be beneficial, provided the individual’s desire for control is balanced against society’s need for safety and the business community’s demand for functionality.  

Rod Garratt is a professor of economics at the University of California, Santa Barbara. 

Michael Junho Lee is a financial research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.   

 
@article{GarrattLee2025,
    author={Garratt, Rod and Lee, Michael Junho},
    title={The Future of Payment Infrastructure Could Be Permissionless},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={November 25},
    year={2025},
    url={https://doi.org/10.59576/lse.20251125}
}

You may also be interested in:

A Historical Perspective on Stablecoins

The Financial Stability Implications of Tokenized Investment Funds

Stablecoins and Crypto Shocks: An Update

There has been a lot of interest in firms’ pricing decisions in the past few years—both during the inflation surge of 2021-23 and in the more recent rounds of tariff increases. In this post, we let firms speak for themselves about what factors they consider when adjusting prices in response to various shocks. The analysis is based on an ongoing research project, joint with the Atlanta and Cleveland Federal Reserve Banks, on how businesses set prices and the extent of passthrough of cost increases. In particular, we leverage the qualitative portion of the study based on open-ended interviews with senior decision-makers on how they approach pricing decisions in their firms. Rather than a uniform approach, a very nuanced picture emerges of businesses trying to balance competing objectives while keeping an eye on demand conditions for their products as well as on their direct competitors’ behavior in the market.

Businesses Trade Off Profit Margins and Sales Volume in Deciding Whether to Pass Through Cost Increases

As described in a previous Liberty Street Economics post, we adopted a two-stage approach to study firms’ pricing behavior, using both qualitative and quantitative methods. Here we focus on the qualitative part of the analysis, based on open-ended interviews with thirty-three business decision-makers (CEOs, CFOs, and business owners) from a variety of industries in 2021 to talk about their price-setting behavior. The second stage of the analysis consisted of a quantitative survey of about 700 businesses fielded in late 2022 and early 2023 across the Second, Fourth and Sixth Federal Reserve districts. The overall findings of the project so far are discussed here. 

Virtually all firms in our sample have a target margin for their operations overall. This is usually expressed as a percent target, although a couple of firms (commodity producers and retailers) target a fixed dollar margin. But even though having an explicit target margin is widespread, this is generally just a benchmark. Almost none of our interviewees mechanically follow a strict cost-plus pricing rule in all circumstances: instead, many factors influence the actual margin a given business is able to achieve.  

One key consideration for firms in deciding whether (and to what extent) to pass through cost increases to their customers is the trade-off between margins and sales volumes. Often, a business will accept a lower margin to protect volume. For example, a fluff and pulp manufacturer stated:  

As an organization, we’re tasked with getting as much as we can from a contribution margin perspective for every time and each amount of time allotted to that machine. But also … there’s risk/reward. You can’t push that too high, because you’ll sacrifice volume.

Along similar lines, an equipment manufacturer stated that their business would only increase prices by 2.5 percent following cost increases of 4 percent, and explained that decision as follows:  

We have a lot of pricing power, but there is still a point where somebody’s like, ‘Hey, I bought one of these two years ago and it’s 15 percent more expensive now. What’s going on?’ We do get … a little bit of pushback. We’re a bit thoughtful about trying to go after too much too fast. 

Conversely, one construction company explained that they might raise margins above their 5 percent target if they have already met their annual sales goal:  

You know what’s going to drive our margins up? When we just don’t feel like we don’t need to add any more work, right? If we get six months through this year and that $600 million that we had in front of us is already converted, just because we can take a chance on not winning another piece of business, we’re going to take some of that 5 percent work and make it 7 percent. 

Demand Conditions Affect the Extent of Pass-Through

As the previous quotes suggest, the trade-off between margins and sales and the extent of cost pass-through are heavily influenced by the demand conditions faced by a business. Most of our interviews were conducted during a period of generally strong demand as the economy emerged from the COVID pandemic. This strong demand made it easier for firms to maintain or even increase margins in response to higher costs. As a residential real estate developer put it:  

So, and again, the good news is that [the] market has allowed us to raise prices, in effect, more than what the cost increases have gone up. But it isn’t necessarily so. It’s not kind of, the cause and effect isn’t, ‘Hey, our costs have gone up so let’s raise prices.’ You know, I mean, if the market demand wasn’t strong enough to raise prices, we’d be stuck with lower margins.

Perhaps not surprisingly, firms are generally aware that demand for their products is downward-sloping and take into account demand elasticities in their pricing decisions. Indeed, some firms explicitly experiment to get a better sense of the slope of the demand curve they face. One producer of dairy ingredients said:  

We’re always constantly trying to push the envelope to see what the market will bear … until you start losing a few [customers], you haven’t achieved the maximum price in the marketplace.  

And one retailer told us: 

We have done some testing … where we have pushed the upper limit in one or two stores … Where we took, basically, a 7 percent increase all at once, to see what the consumer would do, and they retracted.  

Interestingly though, even firms who referred to the concept of “elasticities” do not seem to use them to set markups over costs, as standard theory would suggest: rather, a high elasticity is cited as a reason to refrain from passing through an increase in cost. 

Businesses Watch Their Competitors Carefully 

Another important determinant of firms’ pricing decisions is the behavior of their competitors. One tire manufacturer said:  

The biggest consideration here has been what is the competition doing.… We want to remain price relevant in the market, and understand how much of true cost can we mitigate, or manage, but really it’s what is the competition doing.… If the competition is going 2 percent we don’t want to go out and say, ‘well we really need to pass on four [percent]’ because it won’t hold in the market.  

Another manufacturer put it quite poetically: 

It’s an industry where competition watches each other … it’s a very fine balance and a well-orchestrated dance that we watch each other very closely.  

And a gas station owner said succinctly:  

It’s the competition, whatever is going on in the local market. We try to maximize our profits and protect our volumes.  

This mutual watching among competitors within an industry generates a degree of “strategic complementarity” among businesses: In other words, firms may be more likely to “cross the street in unison” when making pricing decisions. This dynamic may amplify price fluctuations within an industry as firms move in tandem with their competitors, especially when they face common shocks; it may also dampen price fluctuations when firms are hit by shocks that are more firm-specific. 

Firms View Labor Cost Increases Differently than Other Input Cost Shocks 

Finally, an interesting asymmetry in pass-through was revealed in our interviews regarding labor cost increases relative to other cost increases. A sizable share of interviewees (roughly one quarter, all of them in goods producing firms) stated that labor cost increases do not have a first order impact on price increases for their products. Rather, they typically try to offset those labor cost increases through productivity gains and other efficiencies in their production processes. One respondent said:  

We’ve been able to maintain quality, lower our costs through … improvements in our operational capabilities. That’s typically how we think about capturing wage growth, is that if our wages are increasing at three percent annually, we should be aiming for a five percent in sort of efficiency improvements.  

Similarly, a steel manufacturer told us:  

Labor cost is going up. There are things that we have to do to get more efficient, to try to offset that.  

The muted response of these firms’ prices to labor cost increases is consistent with other recent research: “The Missing Inflation Puzzle: The Role of the Wage-Price Pass-Through,” and “The Changing Link Between Labor Cost and Price Inflation in the United States.” In contrast, firms are much more likely to pass through raw materials costs, often via contractual indexation: for a packaging business, “the only adjustments that are made during a contract in terms of pricing has to do with the increases or decreases in raw materials from some baseline that was established at the outset of the beginning of the contract.” 

Conclusion 

Our ongoing project has uncovered some fascinating nuances in how businesses approach their pricing decisions. Virtually all firms have a margin target, but that target is an “aspirational” one: In practice, most firms set margins flexibly and trade off profit margins against sales volume and market share. Demand conditions, and the elasticity of demand for their products are a key factor in price setting and in the extent of pass-through of any cost increases: A stronger demand allows for a higher degree of pass-through, and vice versa. Competitors’ behavior also features prominently in a firm’s calculus regarding pricing: To the extent that businesses in an industry “watch each other carefully,” this has implications for the extent of pass-through and can potentially amplify price fluctuations as “price followers” catch up with “price leaders.” Finally, many firms—at least in goods producing sectors—are less willing to pass through labor cost increases than other input cost shocks, which may limit the passthrough of wage pressures to price inflation. 

Wändi Bruine de Bruin is provost professor of public policy, psychology, and behavioral science at the Sol Price School of Public Policy at the University of Southern California (USC), and director of the USC Behavioral Science and Well-Being Policy initiative.

Keshav Dogra is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Sebastian Heise is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

Edward S. Knotek II is a senior vice president and director of research in the Research Department at the Federal Reserve Bank of Cleveland.

Brent H. Meyer is an assistant vice president and economist in the research department at the Federal Reserve Bank of Atlanta.

Robert W. Rich is the director of the Center for Inflation Research and a senior economic and policy advisor in the Research Department at the Federal Reserve Bank of Cleveland. 

Raphael S. Schoenle is a professor of economics at Brandeis University. 

Giorgio Topa is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Wilbert van der Klaauw is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{BruinedeBruinDograHeiseKnotekIIMeyerRichSchoenleTopavanderKlaauw2025,
    author={Bruine de Bruin, Wändi and Dogra, Keshav and Heise, Sebastian and Knotek II, Edward S. and Meyer, Brent H. and Rich, Robert W. and Schoenle, Raphael S. and Topa, Giorgio and van der Klaauw, Wilbert},
    title={How Businesses Set Prices—In Their Own Words},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={November 24},
    year={2025},
    url={https://doi.org/10.59576/lse.20251124}
}

You may also be interested in:

How Do Firms Adjust Prices in a High Inflation Environment?

Estimates of Cost-Price Passthrough from Business Survey Data

A 2012 Liberty Street Economics post noted that U.S. monetary policy exhibits a surprising degree of seasonal behavior: over the 1987-2008 period, the Federal Reserve was much more likely to lower interest rates (or abstain from raising rates) in the first month of each quarter than in the two subsequent months. Thirteen years later, we revisit that analysis to investigate whether the seasonal pattern in monetary policy still holds today, in the wake of a rate hiking cycle, a pandemic, a surge in inflation, and a second round of rate hikes. We find that the pattern has indeed continued; however, unlike in the earlier sample period, it can be completely explained by the timing of the FOMC calendar.

A Seasonal Pattern in FOMC Decisions

In this post, we focus on a twenty-five-year sample period: January 2000 to July 2025. Note that before October 2008—the endpoint of the period covered in the previous post—the FOMC conducted monetary policy by announcing a target for the federal funds rate (FFR or funds rate). After October 2008, the FOMC conducted policy by communicating a target range for the funds rate.  

As shown in the chart below, since 2008, the funds rate spent six years at the zero lower bound, before a moderate hiking cycle in the late 2010s, another return to the zero lower bound during the COVID pandemic, and a much steeper hiking cycle starting in 2022. Whereas the funds rate declined by around 600 basis points over the twenty-five-year period studied in the 2012 blog post, it only declined, on net, by 125-150 basis points since 2000. 

Monetary Policy Easing and Tightening Cycles over the Past 25 Years

Despite the notable events that have occurred since 2012, surprisingly, we find that changes in the funds rate have exhibited essentially the same seasonal behavior over the past twenty-five years as they did over the 1987-2008 period. This pattern is illustrated in the chart below, which shows the upward and downward changes in the target federal funds rate made at each meeting since the late 1980s (we include unscheduled meetings with rate changes but exclude meetings with no change in the target rate).  

We can see that meetings held in the first month of a quarter (blue dots) are over-represented when there are decreases in the target rate (i.e., negative values in the chart) and under-represented when there are increases in the target rate. Furthermore, this pattern broadly persists after October 2012, when the original blog post was published (marked by a vertical line). Most strikingly, there are many more red dots than blue dots in the top right area of the chart—recent tightening of policy almost always occurred in either the second or third month of the quarter. 

Changes in the Federal Funds Target: 1st Month of a Quarter versus 2nd/3rd Months

We can make this point more precisely by accumulating these changes for each calendar year. The table below shows the total percentage point change in the federal funds rate target for FOMC meetings in the first month of the quarter (January, April, July, and October), compared with the total change over the remaining eight months of the year (we omit years in which there was no change in the target funds rate). The funds rate declined by 4 percentage points, in total, during meetings in the first month of each quarter since 2000. During meetings in the remaining eight months of each year, the funds rate increased by almost 3 percentage points. Since the original blog post was published in 2012, the funds rate increased by 50 basis points in the first month of the quarter but rose by 375 basis points in all other months. Thus, the seasonal pattern appears to be intact. 

Annual Changes in the Federal Funds Target: 1st Month versus 2nd/3rd Months 

As discussed in the original post, this surprising result could be a matter of coincidence, with the schedule of FOMC meetings falling into a certain alignment with the timing of hiking and easing cycles. Over the 1987-2008 period, there happened to be more FOMC meetings in the first month of each quarter during periods when the fed funds target was falling, compared to periods when the rate was rising. For example, in 2005—during the 2004-2006 tightening cycle—there were no FOMC meetings held in the first month of the quarter. The prior blog post, however, found that removing years in which there were particularly few or many FOMC meetings scheduled in the first month of each quarter did not eliminate the statistical and economic significance of the seasonal differences.  

Here, we investigate this issue in a slightly different way. For each year, we calculate the total change in the funds rate. We then imagine a counterfactual world in which this total change was distributed evenly across the FOMC meetings that occurred in that year, taking the schedule of meetings as given (we include unscheduled meetings where a change in the target funds rate occurred). The results are shown in the table below.

Counterfactual Annual Changes in the Federal Funds Target:
1st Month versus 2nd/3rd Months

In this counterfactual world, the totality of rate changes over 2000-2024 resulted in a 306 basis point decline in the first month of the quarter and a rise of 193 basis points in all other months. This is a differential of about 5 percentage points, as compared to 6.88 percentage points in the table above. Thus, while the timing of FOMC meetings can account for a significant fraction of the seasonality observed over this whole period, there is still almost 200 basis points that is not explained by timing.  

In contrast, since 2012 the counterfactual path of changes in the funds rate is almost exactly the same as the historical path shown in the table above. In reality, there was a total increase of 50 basis points in the first month of the quarter as compared to an increase of 57 basis points in the counterfactual exercise. In all other months there was a rise of 3.75 percentage points as compared to 3.67 percentage points in the counterfactual exercise.  This implies that the seasonal pattern we have observed since 2012 is essentially explained in full by the timing of FOMC meetings alone. Thus, under further scrutiny, we can conclude that the “excess seasonality” noted in the original post does not appear to have persisted beyond 2012. 

Richard K. Crump is a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Keshav Dogra is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Dennis Kongoli is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{CrumpDograKongoli2025,
    author={Crump, Richard and Dogra, Keshav and Kongoli, Dennis},
    title={Is Monetary Policy Still Seasonal? },
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={November 19},
    year={2025},
    url={https://doi.org/10.59576/lse.20251119}
}

You may also be interested in:

Is U.S. Monetary Policy Cyclical?

Tracking Reserve Ampleness in Real Time Using Reserve Demand Elasticity

The Central Banking Beauty Contest

Bank failures and distress can be costly to the economy, causing losses to creditors and reducing the flow of credit and other financial intermediation services. Thus, there is significant value in being able to identify “at risk” banks in a timely and accurate way. In a previous post, we presented a new solvency metric, Economic Capital, and showed how solvency risks in the U.S. banking industry have evolved over time according to this measure. In this post, we continue to draw on our recent Staff Report to present analysis showing that Economic Capital identifies failing banks earlier and more accurately than more conventional solvency measures.

An Overview of Economic Capital 

As an alternative to more traditional measures of bank solvency such as regulatory capital and tangible common equity (TCE), we propose and develop a measure of economic capital (EC). EC is calculated using estimates of the present value of bank assets, liabilities, and necessary operating expense. EC embeds changes in the value of bank assets and liabilities due to changes in interest rates and credit spreads over time, in contrast to more traditional solvency measures based on accounting principles, which do not consistently recognize changes in value before contractual maturity. We calculate EC using publicly available regulatory data for nearly all U.S. commercial banks over a long historical period from 1997 to the beginning of 2025.  

As discussed in our prior post, EC has several conceptual and computational advantages relative to measures based on accounting principles. Of particular interest, we can calculate R-EC, or the value of EC under the assumption that uninsured depositors will run on the bank. This measure provides insight into whether banks would continue to be solvent in such circumstances.  

Better Identification of Failing Banks? 

While EC has certain conceptual advantages, does it really do a better job of identifying failing banks earlier or more accurately than accounting-based solvency measures? To address this question, we consider failures during the March 2023 period, when sharp increases in interest rates decreased the value of bank assets, leading to runs and the failure of four large banks. We examine all failing banks between 1997 and 2025, a sample that is dominated by credit-related failures during the 2007-2009 Global Financial Crisis (GFC).  

Test Case: March 2023 Banking Stress Episode 

The chart below tracks four solvency measures in the five years before March 2023. The four measures are our baseline EC measure, R-EC, TCE, and TCE adjusted for estimated losses on loans and securities (MTM TCE). Each measure is scaled by total bank assets to create a leverage ratio-type metric. The chart shows the path of each measure for the four banks that failed during this episode—Silicon Valley Bank (SVB), First Republic, Silvergate, and Signature Bank—along with the 5th to 95th percentile range of each measure for all banks in our sample (“the industry”).  

As the top left panel shows, while the failing banks have low EC ratios, they were not stark outliers with respect to the overall distribution of banks until interest rates began to rise in 2022. In contrast, R-EC ratios for these banks are notably low—sometimes below 5th percentile values—as far back as five years before the industry became stressed in 2023. In fact, the R-EC ratios for SVB and First Republic were negative by mid-2022.  

Overall, R-EC signaled a high risk of insolvency at these banks under a run scenario well ahead of actual events. In contrast, neither of the accounting-based TCE measures provided as clear a signal. While TCE ratios for the failing banks were below industry averages, they were within the 5th–95th percentile range for the entire period. In fact, MTM TCE ratios for these firms actually rose in the distribution—for Silvergate and Signature Bank, to levels above the industry average—over the course of 2022. While the level of MTM TCE ratios for these firms fell in absolute terms, they fell by less than the industry average, illustrating the shortcomings of marking only one side of the balance sheet to market. 

Failed-Bank Solvency Measures

EC

R-EC

TCE

MTM TCE

Sources: Call Reports; authors’ calculations.  
Notes: The chart shows four solvency measures from 2018:Q1 to 2022:Q4 for the four banks that failed or were liquidated in the March 2023 period of banking industry stress. The solvency measures are economic capital (EC), run economic capital (R-EC), tangible common equity (TCE), and mark-to-market TCE (MTM TCE). The four banks are Silicon Valley Bank (SVB, in red), First Republic (in green), Signature (in orange), and Silvergate (in blue). The grey bars are the 5th to 95th percentile range for all commercial banks with assets greater than $50 million, excluding trust banks.

Is It Just Interest Rate Risk? 

While the analysis thus far suggests that R-EC did a better job of anticipating which banks would fail in 2023, this finding may reflect the characteristics of that particular episode rather than a broader capability to accurately identify failing banks. To address this concern, we consider all banks that failed at any time during our sample period, based on information from the FDIC. The resulting sample of 465 failed banks is dominated by failures during the GFC, as roughly two-thirds of the failures occurred from 2008 to 2010, largely reflecting credit losses—distinct from the interest-rate-driven market value losses that spurred the 2023 episode.  

To explore how well the various solvency measures predict the broader set of bank failures, we estimate simple logit models, using a zero-one indicator of bank failure and values of R-EC, TCE, and MTM TCE, in turn, as the explanatory variable. (We do not report results for EC as they are very close to those for R-EC.) We estimate these models using solvency measures eight and twelve quarters prior to failure and use the resulting coefficients to construct “Receiver Operating Curves” (ROCs) showing the tradeoff between correctly identifying a failing bank (on the y-axis) and incorrectly identifying a non-failing bank as failing (on the x-axis). An ideal measure would correctly identify a high portion of failing banks but incorrectly identify only a small portion of non-failing banks, so curves that are sharply sloped and closer to the y-axis (with more area under the curve) indicate better performance of the metric in optimizing this tradeoff. 

The chart below shows the ROCs for the three solvency metrics eight and twelve quarters before failure. At both horizons, R-EC is more accurate than the alternative solvency metrics, with differences at the eight-quarter horizon being particularly notable. In all cases, results for R-EC (the red lines) are steeper and closer to the y-axis than results for TCE and MTM TCE, which perform similarly to one another.  

Receiver Operating Curves for Solvency Measures

8-quarter

12-quarter

Source: Authors’ calculations.
Notes: The chart shows the receiver operating curves (ROCs) distinguishing banks that fail (on the y-axis) from banks that do not fail (x-axis) based on logit regressions using three solvency measures: run economic capital (R-EC), tangible common equity (TCE), and mark-to-market TCE (MTM TCE). The panels show the ROCs 8 quarters and 12 quarters before failure.

Summing Up 

In this post, we present a range of results showing that our measure of Economic Capital is a more timely and accurate indicator of bank failure than more conventional accounting-based measures of bank solvency. Accounting-based measures are founded on the assumption that the bank will continue as a going concern and thus that the contractual timing of payments on assets and liabilities will prevail. Conceptually, this assumption does not capture circumstances in which depositors choose to withdraw their funds and thus is unlikely to provide strong signals about solvency under such circumstances. Economic Capital, in contrast, can be calculated under a bank run scenario to illustrate precisely this form of solvency risk. Our analysis supports the idea that this matters in practice by accurately identifying failing banks, indicating that Economic Capital could be a useful supplement to existing solvency metrics in monitoring both the health of individual banks and the risks facing the banking system. 

Beverly Hirtle is a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.  

At the time this post was written, Matthew C. Plosser was a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{HirtlePlosser2025,
    author={Hirtle, Beverly and Plosser, Matthew C.},
    title={Economic Capital: A Better Measure of Bank Failure?},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={November 6},
    year={2025},
    url={https://doi.org/10.59576/lse.20251106}
}

You may also be interested in:

Economic Capital: A New Measure of Bank Solvency 

Measuring the Financial Stability Real Interest Rate, r**

Mitigating the Risk of Runs on Uninsured Deposits: The Minimum Balance at Risk

Every state in the country has a law permitting involuntary hospitalization if a person presents a danger to themselves or others as a result of mental illness. If a person reaches this high bar, the logic goes, they should be confined in a psychiatric hospital for treatment until they are stabilized. (The process is also sometimes called involuntary commitment, involuntary psychiatric hold, or sectioning.) Although there is no definitive national accounting, it is estimated that about 1.2 million involuntary psychiatric hospitalizations occur every year (Lee and Cohen 2021). This puts the magnitude on par with the 1.2 million individuals imprisoned in state, federal, and military prisons every year (Carson 2022). In a new Staff Report, we use data from Allegheny County, which includes Pittsburgh, to measure how psychiatric commitments are impacting an individual’s risk of danger to themselves or others, earnings, and housing.

In the abstract, the impact of an involuntary hospitalization is unclear. During episodes of mental health crisis, involuntary hospitalization may help remove an individual from a dangerous setting and give them access to stabilizing care. This care can in part connect a person to treatment and/or services outside the hospital. Incapacitation may also reduce risk, since individuals are held in inpatient care under supervision and the majority of both violent offenses and suicides are premeditated very briefly (Brouwers Appelo and Oei 2010; Deisenhammer et al. 2009).

On the other hand, involuntary hospitalization may disrupt beneficial social supports such as therapeutic relationships, housing, and employment. Moreover, if individuals find involuntary hospitalization unwelcome, it may degrade trust in the healthcare system, making it harder to access treatment thereafter.

Variance in Physician Behavior

For involuntary hospitalization to proceed, a physician needs to determine that the individual poses a danger to themselves or to others. This evaluation is performed by a physician on staff in the emergency department. We find that once a person has been brought to a given hospital, in a given shift, which physician performs the evaluation is as good as random. This is because there is a very short window in which the evaluation must occur and so the next available physician takes the case. Which physician evaluates a given case is important because physicians vary substantially in what share of cases they refer for involuntary hospitalization. The chart below shows the distribution of physicians’ tendency to involuntarily hospitalize the patients they evaluate. The vertical lines show the 5th and 95th percentile, revealing that physicians at the lower end of the distribution hospitalize as few as 64 percent of the patients they evaluate while those at the upper end of the distribution hospitalize as many as 93 percent.

Distribution of Physicians’ Tendencies to Involuntarily Hospitalize

We leverage this quasi-random assignment to which physician performs the examination in instrumental variables analysis akin to an examiner research design. In a randomized experiment, a patient is randomly assigned to a treatment or control group. In our context, a patient is quasi-randomly assigned a physician for an exam, and that physician may have a high or low tendency to hospitalize patients. We use this variation in examiner behavior to untangle causal effects. By comparing the outcomes among individuals seen by physicians who hospitalize relatively more of their patients to the outcomes among individuals evaluated by physicians who hospitalize fewer of their patients, we can assess the impact of involuntary hospitalization.

The group that could end up in either treatment or control, depending on which physician assesses them, are called the “compliers”—individuals who are, from the perspective of the evaluating physicians, a judgment call for an involuntary hospitalization. We estimate that roughly 43 percent of those evaluated for involuntary hospitalization fall into this group. The result of any instrumental variables analysis only applies to this “complier” group, the judgment call cases.

Danger to Self and Others

The chart below shows the local average treatment effect (LATE) of hospitalization on being charged with a violent crime or dying by suicide or overdose in the six months following an evaluation.

We find that in these judgment call cases, involuntary hospitalization significantly increases the likelihood of harm to self or harm to others. In particular, for judgment call cases we find that the risk of a being charged with a violent crime in the three months following an evaluation is increased by 2.6 percentage points above a baseline of 3.3 percent  (though the chart shows results for each month through the six months after evaluation). Likewise, for judgment call cases, involuntary hospitalization increases risk of dying from suicide or drug overdose death is increased by 1.0 percentage point above a baseline risk of 1.1 percent over a three-month period after evaluation for hospitalization.

Increases in Charges for Violent Crime and Death by Suicide/Overdose for Judgment Call Cases 

This result is surprising. Involuntary hospitalizations are a public safety measure, and the finding that they are driving more of the outcomes they seek to prevent in the judgment call subpopulation we study has important policy implications. The significance is especially pronounced since many areas across the country are seeking to scale up involuntary hospitalizations.

Interpretation

Why might involuntary psychiatric hospitalization make someone more likely to die by suicide or drug overdose or be charged with a violent crime? To better understand our main results, we assess whether involuntary hospitalization affects an individual’s earnings and housing status. Using the same instrumental variables approach, we see that earnings drop significantly for those in the “complier” (judgment call) group who are hospitalized. We also see significantly more homeless shelter usage for people who have not used shelter before and are among these judgment call cases, indicating a destabilization of housing.

We do not observe significant improvements in medication adherence or engagement with outpatient care in the months after the judgment call evaluations, indicating that involuntary commitment is not significantly connecting judgment call individuals to care.

This evidence together suggests that, on net, the destabilizing forces are more powerful than the therapeutic ones for the “complier” (judgment call) group we assess in this study. This connects with prior evidence that destabilizing forces can increase the likelihood of adverse outcomes (Lin 2008; Sullivan and von Wachter 2009; Eliason and Storrie 2009; Dobkin et al. 2018). It also connects to related research documenting that defensive medicine can have adverse consequences for patients’ wellbeing (Kessler and McClellan 1996;  Studdert et al. 2005).

The result has broad policy implications. Public cases in which persons who have not been involuntarily hospitalized and have subsequently engaged in violent behavior have prompted calls for expansions of involuntary hospitalization (Hirschauer 2025). Our results suggest that, if involuntary hospitalization systems in other areas have similar effects on patients to those we document in our study, it may be worth exploring additional or alternative measures to support individuals in mental health crises. 

Moreover, our analysis suggests multiple lines of future research. Outcomes among those evaluated for involuntary hospitalization are very poor, whether they are hospitalized or not (Welle et al. 2023), suggesting a need to develop better forms of care for people facing psychiatric emergencies. The more we understand when involuntary hospitalization is likely to improve patient outcomes and when it is likely to hurt outcomes, the better targeted this intervention can be.

Lastly, from a fairness lens more work should be done to assist physicians in their decision-making processes and to reduce the variance across physicians in the tendency to hospitalize. Better utilization of scarce healthcare resources, including emergency and inpatient hospital beds, has the potential to improve care for all.

Natalia Emanuel is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Pim Welle is chief data scientist for the Allegheny County Department of Human Services in Pittsburgh, PA.

Valentin Bolotnyy is a Kleinheinz Fellow at Stanford University’s Hoover Institution.

 
@article{EmanuelWelleBolotnyy2025,
    author={Emanuel, Natalia and Welle, Pim and Bolotnyy, Valentin},
    title={A Danger to Self and Others: Consequences of Involuntary Hospitalization},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={October 15},
    year={2025},
    url={https://doi.org/10.59576/lse.20251015}
}

You may also be interested in:

A Danger to Self and Others: Health and Criminal Consequences of Involuntary Hospitalization

In a 2024 post, we showed that interdealer trading in benchmark U.S. Treasury notes and bonds concentrates on the last trading day of the month, likely due to passive investment funds’ turn-of-month portfolio rebalancing. In this post, we extend our trading activity analysis to the full range of Treasury securities and market segments. We find that trading is even more concentrated on the last trading day of the month for other types of Treasury securities and in the dealer-to-customer segment of the market, with trading volume in off-the-run Treasuries twice as high as on other days, on average. 

Turning from Interdealer Data to TRACE Data 

In our earlier post, we showed that interdealer trading volume in benchmark (or on-the-run) Treasury notes and bonds is roughly 46 percent higher on the last trading day of the month.  Our data for that work came from the interdealer broker market, in which dealers and principal trading firms transact. Our data source did not include trading by customers and it did not include trading in off-the-run securities. 

In this post, we use data from the Financial Industry Regulatory Authority’s (FINRA) Trade Reporting and Compliance Engine (TRACE). TRACE captures data on all trades of dealers and certain depository institutions, including their trades with customers. Moreover, TRACE covers trades across the full range of Treasury securities, including bills, Treasury Inflation-Protected Securities (TIPS), floating rate notes (FRNs), and both off-the-run and on-the-run notes and bonds. 

Our source of Treasury TRACE data is FINRA’s Treasury Daily Aggregate Statistics, with daily information available from February 13, 2023. The data are aggregated across dealers and across broad categories of securities for all Treasuries except on-the-run notes, bonds, and TIPS, for which activity is reported by security. Activity is further split into “ATS (Alternative Trading System) and Interdealer” (which we refer to as “Interdealer”) and “Dealer-to-Customer” segments. 

Controlling for Day-of-Week Effects 

As in our earlier post, we control for day-of-week effects in our analysis. This matters because the last trading day of the month is about three times more likely to fall on a Friday than on other weekdays and because trading volume tends to be about 11 percent lower on Fridays, on average. By comparison, volume is 9 percent lower than average on Mondays, whereas volume is higher than average on Tuesdays, Wednesdays, and Thursdays (by 3 percent, 9 percent, and 6 percent, respectively). 

End-of-Month Patterns Are Stronger Than Prior Work Suggests 

Overall Treasury security trading volume is about 58 percent higher on the last trading day of the month than on other days, on average (see chart below). This increase is appreciably larger than the 46 percent increase found in our earlier blog post, which examined interdealer broker trading of on-the-run notes and bonds. These findings imply that the end-of-month pattern is even stronger in the dealer-to-customer segment and/or for off-the-run securities.  

End-of-Month Patterns Vary by Market Segment 

The end-of-month increase in trading volume in the dealer-to-customer segment is 73 percent, as shown in the chart below, much greater than the 46 percent increase observed in the interdealer segment. The larger increase in the dealer-to-customer segment does not seem surprising given our conjecture that the end-of-month pattern is driven by customer activity. Dealers likely offset some but not all of their trading with customers in the interdealer segment. 

End-of-Month Patterns Vary by Security Type 

The end-of-month increase in trading volume also varies across securities, as shown in the next chart, averaging 220 percent for FRNs, 167 percent for TIPS, 63 percent for notes and bonds, and 30 percent for bills. The larger end-of-month increases for FRNs and TIPS may reflect these securities’ more common end-of-month issuance and maturity dates as well as investors’ greater inclination to concentrate their trading in these less active securities. For context, daily trading volume over our sample period averages about $3 billion for FRNs, $18 billion for TIPS, $174 billion for bills, and $701 billion for notes and bonds. 

The table below shows that the patterns discussed above generally hold across security types within a market segment and across market segments within a security type. The one exception is that we find a larger end-of-month increase for FRNs in the interdealer segment than the dealer-to-customer segment. This is explained by the fact that the FRNs’ interdealer segment has low volume generally, with some large outliers for the end-of-month increases. So when looking at the median end-of-month increase, which is less sensitive to outliers, the interdealer segment for FRNs has a smaller increase than the dealer-to-customer segment. 

End-of-Month Patterns Vary by On-the-Run/Off-the-Run Status 

Our last chart shows that the end-of-month increase in trading volume is 108 percent for off-the-run securities versus 52 percent for on-the-run securities. The increases are larger for TIPS than for notes and bonds, as shown above, with the on-the-run/off-the-run differentials similar across security types. The larger end-of-month increases for off-the-runs may reflect customers’ greater ownership and hence trading of these securities. Trading in on-the-run securities, in contrast, is concentrated among dealers and principal trading firms. 

The table below shows that the patterns discussed above hold within a particular market segment, security type, and on-the-run/off-the-run status. That is, the end-of-month volume increase is more pronounced in the dealer-to-customer segment, for TIPS, and for off-the-run securities. 

Summing Up 

Trading activity is more highly concentrated on the last trading day of the month than our earlier analysis suggests. In particular, the end-of-month increase in trading volume is greater in the dealer-to-customer segment, greater for off-the-run securities, and greater for FRNs and TIPS than for notes and bonds. These findings are important for market participants managing trade execution strategies and for policymakers monitoring market functioning and liquidity provision. Future work could explore whether the increased end-of-month trading in these other parts of the market is associated with improved liquidity, as we found for benchmark notes and bonds in our past work. 

Michael J. Fleming is head of Capital Markets in the Federal Reserve Bank of New York’s Research and Statistics Group. 

Jonathan Palash-Mizner, a former undergraduate intern at the Bank, is a student at Yale University.

Or Shachar is a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{FlemingPalash-MiznerShachar2025,
    author={Fleming, Michael J. and Palash-Mizner, Jonathan and Shachar, Or},
    title={End‑of‑Month Activity Across the Treasury Market},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={October 9},
    year={2025},
    url={https://doi.org/10.59576/lse.20251009}
}

You may also be interested in:

The Fed’s Treasury Purchase Prices During the Pandemic

Has Treasury Market Liquidity Improved in 2024?

Are There Too Many Ways to Clear and Settle Secured Financing Transactions?

Over the past few months, New Jersey and Vermont have joined a growing number of U.S. states in requiring employers to include an estimated salary range in their online job listings. Has this push for greater pay transparency been effective? In this post, we use granular data on U.S. job postings from Lightcast to assess employers’ compliance with these new regulations. Focusing on the jurisdictions that adopted pay transparency laws early on, we find that many employers ignore pay transparency requirements; roughly a quarter of job listings covered by these laws fail to include salary information.

Aggregate Pay Transparency in Online Job Ads 

Across the United States, the monthly share of online job postings with pay information has risen dramatically, from an average of 15 percent before January 2018 to approximately 53 percent since January 2024. In the Lightcast data, pay information is recorded in the form of an annualized upper and lower bound of advertised salary. For most jobs posted in recent years, pay information is an actual range. The share of job postings with the same lower and upper bounds (a “point salary”) has plateaued at about 15 percent, as shown by the red line in the chart below. While researchers have suggested potential explanations for this overall trend, notably the inclusion of an estimated salary range by some major job listing platforms, we focus on recent legislation requiring employers to include an expected starting pay range in their job listings. How has pay transparency in online job ads evolved in the jurisdictions that have enacted these requirements?  

Pay Transparency Requirements 

To date, ten states, the District of Columbia, and multiple local jurisdictions have adopted such pay transparency laws; the most recent states to do so—New Jersey and Vermont—began implementation in June and July of 2025, respectively. Broadly, these laws require firms that advertise jobs externally to include a “good faith” estimate of the salary or wage range that they expect to pay. These laws are frequently proposed and written with the aim of addressing pay disparities. For example, Massachusetts’ fact sheet for employers begins by stating that its pay transparency law was signed “to increase equity and transparency in pay in the Commonwealth,” with pay transparency cited as “one of the best tools to close gender and racial wage gaps.” 

In this post, we focus on the three earliest states to implement these laws—Colorado (January 2021), California (January 2023), and Washington (January 2023)—in addition to New York City (November 2022); combined, these four jurisdictions account for about 20 percent of all job postings across the United States.  

The chart below shows the monthly share of job postings with salary information in these four jurisdictions; for comparison, the “other” line tracks the average percentage for all other states excluding New York. We remove non-New York City postings from the New York State data, as the state began implementation of a parallel pay transparency law in September 2023. The dashed lines demarcate when legislation enforcement began. We can see the impact of pay transparency requirements: in the month when regulations were implemented, the share of postings with salary information increased by an average of 20 percentage points, while there was a similar 27 percentage point increase across the entire January 2020-March 2025 period for states without such a law. We can also see that the post-transparency-legislation shares do not reach 100 percent. However, given that businesses with few employees are exempt from pay transparency requirements, to what degree does that shortfall reflect noncompliance? 

Expected versus Observed Compliance 

Exceptions for small employers, temporary help agencies, and narrow cases of remote work imply that employers could be in full compliance even if not all postings include salary information. Specifically, employers with a national headcount of fewer than fifteen workers are excluded from California’s and Washington’s pay transparency requirements. This threshold is even lower for employers in New York City (four) and Colorado (one employee in the state). In the panels below, the light blue lines show the evolution of the monthly share of job postings with salary information for each state, plotted against estimates of the expected share under full compliance.  

We use two supplementary data sources to construct these benchmarks. First, we use the Bureau of Labor Statistics’ Job Openings and Labor Turnover Survey (JOLTS) estimates of job openings by establishment size class to approximate the share of vacancies belonging to establishments with at least ten employees between January 2023 and April 2025. The resulting estimate (80.4 percent) is plotted in the panels below as a red line.  

Observed Compliance with Pay Transparency Laws Falls Short of Compliance Expected from Laws’ Firm Size Exemptions

Sources: Lightcast; Bureau of Labor Statistics, JOLTS; authors’ calculations.
Notes: The Bureau of Labor Statistics does not report establishment size estimates by state, so we assume that this distribution is constant across states. Additionally, the laws’ size restrictions are based on firm (national) employment rather than local establishment sizes. 

For a more refined approach, we turn to firm characteristics identified by Lightcast, which include firm size classes. In the panels above, the gold line represents the average monthly share of Lightcast job postings in each state originating from firms with at least eleven employees. Although there is minor variation across the four states, this share is approximately 87 percent. From the substantial gap between the light blue and gold lines, we can observe that while actual compliance with pay transparency laws is high, it falls short of our back-of-the-envelope estimation of expected compliance.  

Even when breaking out the share of compliant postings by firm size class in the charts below, we fail to observe 100 percent compliance. If anything, the largest firms—which should all be bound by pay disclosure laws—are least likely to voluntarily post salary information in job postings prior to regulations. After pay disclosure laws are implemented, the shares of job postings with salary information from firms in the three largest size classes are in line with the share from firms with ten or fewer employees. Taken together, the online postings data provided by Lightcast suggest that, as of January 2025, approximately 24 percent of ads do not comply with pay transparency requirements. Echoing these findings, several jurisdictions, such as Colorado and New York City, have conducted violation investigations and issued notices or fines since these laws were passed.  

A Substantial Share of Firms Bound by Pay Transparency Laws Remain Noncompliant

Share of postings by firm size (number of employees)

Sources: Lightcast; authors’ calculations.

Conclusion 

Across a range of jurisdictions, we document that pay transparency in job postings remains incomplete several years after legal mandates came into force: nearly a quarter of ads bound by mandates in early adopting states still omit wage information. The reasons for noncompliance, however, are still unclear. Are employers unaware of the new rules requiring pay transparency in job postings or do they actively withhold the expected salary? Tracking compliance only represents the first step toward understanding the response of employers to pay transparency regulations.  

Richard Audoly is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Roshie Xing, a former research analyst at the Federal Reserve Bank of New York, is currently a first-year PhD candidate in economics at Stanford University.

 
@article{AudolyXing2025,
    author={Audoly, Richard and Xing, Roshie},
    title={Do Employers Comply with Pay Transparency Requirements in Job Postings?},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={October 2},
    year={2025},
    url={ https://doi.org/10.59576/lse.20251002}
}

You may also be interested in:

Are Businesses Scaling Back Hiring Due to AI?

A New Indicator of Labor Market Tightness for Predicting Wage Inflation

An Update on the Reservation Wages in the SCE Labor Market Survey

The bank run that started in March 2023 in the U.S. occurred at an unusually rapid pace, suggesting that depositors were surprised by these events. Given that public data revealed bank vulnerabilities as early as 2022:Q1, were other market participants also surprised? In this post, based on a recent paper, we develop a new, high-frequency measure of bank balance sheet risk to examine how stock market investors’ risk sensitivity evolved around the run. We find that stock market investors only became attentive to bank risk after the run and only to the risk of a limited number (less than one-third) of publicly traded banks. Surprisingly, investors seem to have mostly focused on media exposure and not fundamentals when evaluating bank risk. In a companion post, we examine how the Federal Reserve’s liquidity support affected investor risk perceptions.

How Risky Were Banks Before the Bank Run?

We emphasize two balance sheet features that turned out to be particularly problematic during the bank run: the share in total assets of uninsured deposits (denoted UID) and the share in total assets of unrealized losses on securities held in accounts not intended for trading (denoted Losses). High values of UID proved to be risky as they were concentrated in certain sectors, which heightened the risk of rapid withdrawals. When Losses are high (typically when interest rates are increasing, as in 2022) and ultimately realized, bank capital is more likely to be eroded below regulatory limits.

To benchmark bank risk, we construct four groups of publicly traded banks that were differentially affected during the bank run: distressed banks that were downgraded in April 2023, large regional banks (those included in the Regional Banking Index, or KRX) that were at the heart of the crisis, small regional banks (those with assets greater than $10 billion that are not included in any bank index), and stress-tested banks, or STBs (the large banks that participated in the Federal Reserve stress tests of 2022 and that are included in the broad bank index, or KBW).

The chart below shows the median values of UID (left panel) and Losses (right panel) in 2022 by bank group. For reference, we also include the three banks that failed in March 2023 (Silicon Valley Bank, Signature Bank of New York, and Silvergate Bank), although they are not part of our analysis. Bank vulnerabilities were apparent as far back as 2022:Q1. Banks that would later fail or be distressed stood out with the highest levels of UID and Losses. Notably, Losses increased for all bank groups as the Federal Reserve raised rates, peaking in Q3. While large regional banks did not have atypical levels of UID, they had more Losses than smaller regionals and STBs.

Bank Vulnerabilities Were Apparent in 2022

Sources: Federal Reserve Board, Consolidated Financial Statements of Bank Holding Companies (FR Y-9C data); Federal Financial Institutions Examination Council, Consolidated Reports of Condition and Income (Call Reports).
Notes: The chart shows the median asset shares of uninsured deposits (UID) and unrealized securities losses (Losses) in 2022 by groups of publicly traded banks. Failed banks are those that were liquidated or failed in March 2023. Distressed banks are those that were downgraded in April 2023. Large regional banks consist of non-downgraded regional banks in the Regional Banking Index (KRX). Small regional banks are those with assets of at least $10 billion that were not included in any bank index. Stress-tested banks were part of the Federal Reserve’s stress tests in 2022 and part of the broad bank index KBW.

Did Stock Market Investors Monitor Risky Banks?

Our novel measure of bank balance sheet risk is constructed as follows. First, we calculate the average stock returns of a portfolio of banks with a particular balance sheet characteristic (such as UID) in 2022:Q3. For example, the UID portfolio return is the average difference in stock returns of banks with the highest UID minus those with the lowest UID in 2022:Q3. If banks with higher UID are riskier, this factor is expected to have a positive return on average (to compensate stock market investors for the greater risk).

Next, to measure how investors perceived the UID risk of a bank, we estimate the co-movement of a bank’s excess stock returns with the UID portfolio return, after accounting for other types of risk (for example, size, value, and stock market risk). We denote this co-movement as the UID beta. If a bank’s UID beta increases, this indicates that investors are more sensitive to the bank’s systematic UID risk. We construct the Losses beta using an identical procedure. (Our paper also constructs betas with respect to cash and regulatory capital.)

The chart below shows estimates (indicated by the dots) of the average UID beta and Losses beta for all banks before the run (January–February 2023; Pre in the chart) and during the run (March 1–May 5, 2023; Post in the chart). The lines through the dots indicate confidence intervals. The chart shows that, before the run, the betas were not statistically different from zero (since the confidence interval straddles zero). But after the run the betas become positive and statistically significant. In other words, investors mostly ignored risks from high levels of uninsured deposits and unrealized losses on securities—until the crisis actually hit.

Investors Mostly Ignored Bank Risk Until the Bank Run Hit

Which Banks Did Investors Run On?

At the individual bank level, we find that the betas increased during the run for about a third of all publicly traded banks. Thus, investor concerns about bank risk were seemingly not broad-based. What characterized this limited set of banks? Surprisingly, we find that balance sheet variables as of 2022:Q3 or Q4 fail to predict which banks had significantly higher betas during the run. In other words, banks perceived as riskier by investors during the run were seemingly not the ones with worse fundamentals in 2022.

How News Drove Risk Perceptions

If fundamentals did not drive investor attention, then what did? We consider whether news coverage facilitated the coordination of investor attention on certain banks. Such a possibility has previously been found in the context of bank failures. We define a bank’s news coverage as the number of articles about a bank on a given day divided by the bank’s assets, to account for larger banks having more publications. We denote this variable Pubcount. In the chart below, Pubcount (measured as deviations from zero, and in standard deviation units) shows considerable daily variation, implying that even banks with low average media coverage experience periods of relatively intense publicity. Through March 8, just before the crisis, Pubcount was negative (i.e., below average) for all groups. When some distressed banks were put on a downgrade watch by Moody’s after the markets closed on March 13, Pubcount spiked for all distressed banks. Media interest surged again when the distressed banks were downgraded starting on April 14. In general, it appears that increases in Pubcount are associated with risk events.

Bank Publications Increase Around Risk Events

Does news coverage affect the balance sheet betas? In the chart below, we show estimates of news betas (or the component of betas that vary with Pubcount) around the bank run. These news betas are significantly negative before the run (i.e., publications were associated with lower investor risk sensitivity) but became significantly positive during the run (i.e., publications were associated with higher investor risk sensitivity). These effects are economically meaningful as the news betas are at least as large as the non-news betas. The effect of news on the betas persisted for days after publication, suggesting that investors paid attention to news only when it became salient to them.

When a bank appeared in the news during the crisis, investors became much more sensitive to its risks—regardless of whether the bank had a riskier balance sheet than its peers. This result reinforces the notion that news coverage coordinated investors’ actions (and thereby their perceptions of bank risk), either because it reflected latent risk events not captured by balance sheet data, or because investors overreacted to the news.

Final Words

During the bank run of 2023, news flows were at least as important as underlying bank fundamentals in driving investor perceptions of bank risk. News coverage, even when stale, appeared to have served as a coordination device, helping investors focus collectively on certain banks. These results imply that investors may be unable to quickly process information in a crisis, potentially making market price dynamics noisier, to the detriment of market participants and policymakers. However, as investor attention was focused on a few banks rather than a broad swathe of the banking sector, the contagion was contained. Liquidity support by the Federal Reserve may also have limited contagion, a topic we examine in our companion post.

Natalia Fischl-Lanzoni, a former research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group, is pursuing a master’s in computer science at NYU Courant.

Martin Hiti, a former research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group, is a Ph.D. student in finance at the MIT Sloan School of Management.

Asani Sarkar is a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{NataliaFischl-Lanzoni,MartinHiti,andAsaniSarkar2025,
    author={Natalia Fischl-Lanzoni, Martin Hiti, and Asani Sarkar},
    title={Reading the Panic: How Investors Perceived Bank Risk During the 2023 Bank Run},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={September 30},
    year={2025},
    url={https://doi.org/10.59576/lse.20250930a}
}

You may also be interested in:

Calming the Panic: Investor Risk Perceptions and the Federal Reserve’s Emergency Lending During the 2023 Bank Run

Investor Attention to Bank Risk During the Spring 2023 Bank Run

Anatomy of the Bank Runs in March 2023

This post presents an update of the economic forecasts generated by the Federal Reserve Bank of New York’s dynamic stochastic general equilibrium (DSGE) model. We describe very briefly our forecast and its change since June 2025. To summarize, the model expects growth in 2025 to be stronger, and inflation lower, than in June. Moreover, the model’s predictions for the short-run real natural rate of interest (r*) have increased relative to June throughout the forecast horizon, partly reflecting the strength in the economy and the buoyant financial conditions. 

Note: The DSGE model forecast is not an official New York Fed forecast, but only an input to the Research staff’s overall forecasting process. For more information about the model and variables discussed here, see our DSGE model Q & A. 

The New York Fed DSGE model forecasts use data released through 2025:Q2, augmented for 2025:Q3 with the median forecasts for real GDP growth, core PCE inflation, and short-run inflation expectations from the August release of the Philadelphia Fed Survey of Professional Forecasters (SPF), as well as the yields on 10-year Treasury securities and Baa-rated corporate bonds based on 2025:Q3 averages up to August 18. Starting in 2021:Q4, the expected federal funds rate (FFR) between one and six quarters into the future is restricted to equal the corresponding median point forecast from the latest available Survey of Market Expectations (SME) in the corresponding quarter. For the current projection, this is the July SME. 

Growth in 2025 is expected to be stronger, and inflation lower, than in June. This is the result of two interconnected factors. First, the economy turned out to be firmer than anticipated in June for both 2025:Q2 and, according to the SPF projections, 2025:Q3 as well. Also, inflation in these two quarters was lower than the model expected. Second, and possibly related, the effects of tariffs on economic activity and inflation, which the model infers from the 2025 SPF core inflation forecasts, is less than predicted in June. As a reminder, we augmented the model with one- and two-period anticipated cost-push shocks in 2025:Q2 and Q3 to incorporate the effects of tariff announcements on the economy.  

Output growth is higher than predicted in June for 2025 and 2026 (1.4 versus 0.3 percent, and 0.9 versus 0.1 percent, respectively) but slightly lower for 2027 and 2028 (0.8 versus 1.0 percent, and 1.3 versus 1.7 percent, respectively). The lower long-run growth rate is largely a repercussive effect of stronger growth in the short run as negative cost-push shocks, which capture the effects of tariffs, have a level effect on output that dissipates over time according to the model (so that lower growth now implies higher growth later, and vice versa when the shock is positive, as was the case between June and September). 

As mentioned, core PCE inflation is expected to be lower than projected in June for 2025 and 2026 (2.8 versus 3.4, and 1.8 versus 2.1, respectively) and essentially unchanged for 2027 and 2028 (1.7 versus 1.6, and 1.8 versus 1.6 percent, respectively). Uncertainty about both output growth and inflation is considerably lower than it was in June, especially for the current year. 

In terms of assessing the policy stance, the model’s predictions for the short-run real natural rate of interest (r*) have increased noticeably relative to June throughout the forecast horizon (2.6, 2.1, 1.6, and 1.4 percent, respectively, in 2025, 2026, 2027, and 2028 versus 2.2, 1.7, 1.3, and 1.1 percent in the June forecast), partly reflecting the strength of the economy and buoyant financial conditions. The model’s expectations for the policy rate, which are informed by the SME, have not changed much in nominal terms since the beginning of the year, but expected inflation is still higher than it was earlier in the year because of tariffs, resulting in a lower real rate. As a consequence, given the elevated short-run r*, the model views the policy stance as slightly accommodative over the next few quarters.   

Forecast Comparison

Forecasts of Output Growth

Forecasts of Inflation

Real Natural Rate of Interest

 
@article{DelNegroDiagneDograElbarmiLeePham2025,
    author={Del Negro, Marco and Diagne, Ibrahima and Dogra, Keshav and Elbarmi, Elena and Lee, Donggyu and Pham, Michael},
    title={The New York Fed DSGE Model Forecast—September 2025},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={September 19},
    year={2025},
    url={https://libertystreeteconomics.newyorkfed.org/2025/09/the-new-york-fed-dsge-model-forecast-september-2025/}
}

Marco Del Negro is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Ibrahima Diagne is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Keshav Dogra is an economic research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

Elena Elbarmi is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Donggyu Lee is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group.

Michael Pham is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

You may also be interested in:

The New York Fed DSGE Model Forecast—June 2025

Are Financial Markets Good Predictors of R‑Star?

Are Businesses Absorbing the Tariffs or Passing Them On to Their Customers?

Bank supervisors, industry analysts, and academic researchers rely on a range of metrics to track the health of both individual banks and the banking system as a whole. Many of these metrics focus on bank solvency—the likelihood that a bank will be able to repay its obligations and thus retain its funding and continue to supply services to consumers, businesses, and other financial institutions. We draw on our recent research to describe a new solvency metric that is more forward-looking, more timely, and more comprehensive in its assessment of solvency than many current measures. 

Economic Capital: Combining Credit, Liquidity, and Funding Risk 

Arguably the most widely used solvency measures are regulatory capital and tangible common equity (TCE), both of which are grounded in U.S. Generally Accepted Accounting Principles (GAAP). Under U.S. GAAP, many assets and liabilities on banks’ balance sheets are carried at historical or amortized cost rather than at fair market values. While credit losses are recognized via reserving, changes in value due to movements in interest rates and credit spreads are only partially recognized. A key underlying assumption of this approach is that the bank will continue to be a going concern even after facing losses in value, receiving payments due and making payments owed based on contractual obligations.  

However, many bank assets and liabilities do not have binding contractual payment dates, providing bank customers with the option to prepay (in the case of loans) or withdraw funds on demand (in the case of deposits). When a bank’s solvency is in question, depositors have an incentive to run on the institution in an attempt to avoid losses. These incentives are documented in seminal academic work by Diamond and Dybvig (1983). Hence, solvency measures based on accounting standards are not able to quantify the interactions between bank solvency and liquidity that can threaten banks. 

To address these shortcomings, we develop an alternative measure of bank solvency based on estimates of the present values of assets, liabilities, and operating expenses. This measure, Economic Capital (EC), incorporates changes in bank value due to movements in interest rates and credit spreads as well as the changes in value from credit losses recognized in traditional solvency measures. Importantly, EC also incorporates assumptions about the timing of payments and the stability of deposits.   Specifically, we are able to consider cases where uninsured deposits must be replaced with market-rate financing and under potential stress shocks to interest rates and credit spreads.  

Conceptually, EC is closely related to the Economic Value of Equity (EVE), a measure that has long been used to assess banks’ interest rate risk exposures. Traditionally, EVE-based measures have been used to assess changes in economic value (∆EVE) resulting from potential interest rate movements, often scaled by an accounting-based measure of bank capital. This sensitivity is intended to capture capital at risk due to interest rate movements. In contrast, our implementation of EC focuses on the level of economic capital. In this sense, EC is comparable to other solvency measures such as regulatory capital and TCE. By comparing the level of EC to its sensitivity to interest rates, market prices, and depositor behavior, we can quantify banks’ solvency in response to market conditions and funding shocks. 

Calculating Economic Capital 

We calculate EC using publicly available regulatory data for U.S. commercial banks—the Call Reports. These reports contain detailed quarterly balance sheet and income statement information and, critical for our purposes, information about the remaining maturity of each bank’s loan portfolio, time deposits, and long-term debt. The Call Reports also contain bank-reported estimates of the fair market value of securities and trading account assets and liabilities, which we incorporate directly into our measure of EC. Using the information in the Call Reports, we can estimate the present value of loans, demand and time deposits, and other liabilities. As detailed in our paper, our estimates incorporate bank-specific, time-varying factors such as the relative riskiness of the loan portfolio and the price sensitivity of demand deposits to interest rates (“deposit beta”). We also generate bank-specific, time-varying estimates of the minimum operating expenses necessary for each bank to realize the value of its assets and to supply deposit-related services. EC equals our estimates of the present value of assets minus the present value of liabilities minus the present value of these minimally necessary operating expenses.  

 The Call Report data enable us to calculate EC back to 1997, a period that spans several business and interest rate cycles, as well as two periods of significant banking industry stress: the 2007-09 Global Financial Crisis (GFC) and the 2023 episode of banking industry turmoil. This long history provides rich context for assessing the outcomes we produce and for analyzing how the banking industry’s solvency has evolved over time. 

Banking Industry Solvency Over Time 

The chart below presents our baseline measure of EC assuming that banks retain their uninsured demand deposits. This is a benign scenario in that economic value is calculated assuming that banks do not incur funding stress. The chart depicts EC scaled by book assets—an EC leverage ratio—for all U.S. commercial banks between 1997:Q2 and 2025:Q1. The blue diamonds are the unweighted average across all banks, the red diamonds are the average weighted by bank asset size, and the gray shading represents the 5th-95th percentile estimates across all banks.  

On average, EC has varied between 10 and 25 percent of assets for most of the sample period. There was a pronounced dip during the GFC, when banks experienced large credit losses and credit spreads (the black dashed line) widened significantly. During this period, a significant share of banks had negative economic capital. That decline is consistent with concerns that motivated policy interventions to stabilize the banking sector even though conventional solvency measures showed much less deterioration. 

EC has since risen above pre-GFC levels, averaging 20 percent or more of assets since the mid-2010s. Larger banks typically had higher levels of EC than smaller banks during this period (the weighted average exceeds the unweighted average), implying that these banks increased their capital in the wake of the financial crisis. 

Economic Capital (EC) Over Time

The results for our baseline EC measure thus present a relatively encouraging perspective on banking industry solvency. However, because baseline EC is calculated assuming stable funding, it does not incorporate funding liquidity stress. The next chart presents estimates of economic capital under a “run scenario” (R-EC), where we assume that uninsured deposits must be replaced with market-rate funding which is more expensive than deposits.  

Stressing funding costs raises the present value of liabilities by bringing the timing of deposit payments forward in time and assuming they are replaced with more expensive funding sources. The result is that R-EC is less than EC. While R-EC increases after the sharp decline during the GFC, the post-GFC level of R-EC is not materially higher than the level prevailing before the financial crisis. In addition, smaller banks tend to have higher levels of R-EC than larger banks (the unweighted average exceeds the weighted average) due to large banks’ greater reliance on uninsured deposits, suggesting that the typical large bank has higher risk under a run than smaller institutions. Thus, based on our measure of stressed economic capital, post-crisis changes in prudential regulation aimed at larger banks do not appear to have resulted in materially lower solvency risk for these firms, either over time or relative to smaller banks. 

Run Economic Capital (R-EC) Over Time 

The next chart, which plots the difference between EC and R-EC, presents an additional perspective on evolving risks to banking industry solvency. This difference is a measure of the extent of deposit funding risk, as it captures the change in economic capital resulting from a run on uninsured deposits. The banking industry’s exposure to this risk has risen over time, especially for larger banks. The weighted average, unweighted average and cross-bank range have all increased steadily since the GFC, reflecting increased reliance on uninsured deposits in banks’ funding structures.  

Deposit Funding Risk

Summing Up 

In this post, we present new measures of bank solvency that incorporate changes in economic value stemming from movements in interest rates and credit spreads under a range of assumptions about depositor behavior. While accounting and regulatory capital ratios suggest that the banking industry is better capitalized relative to before the GFC, economic capital suggests a more nuanced story. When we assume that deposit funding is stable, it appears the banking industry has improved its solvency since the early 2000s. However, when we assume deposit funding is stressed, we find that economic capital has not materially changed relative to pre-GFC. This reflects the banking industry’s increasing reliance on uninsured deposits as a source of funding, particularly among larger banks. In a future post, we will examine whether our economic capital measures do a better job of identifying failing banks than alternative solvency measures.  

Beverly Hirtle is a financial research advisor in Financial Intermediation Policy Research in the Federal Reserve Bank of New York’s Research and Statistics Group.

Matthew Plosser is a financial research advisor in the Federal Reserve Bank of New York’s Research and Statistics Group.

 
@article{BeverlyHirtleandMatthewPlosser 2025,
    author={Beverly Hirtle and Matthew Plosser },
    title={Economic Capital: A New Measure of Bank Solvency },
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={September 3},
    year={2025},
    url={https://libertystreeteconomics.newyorkfed.org/2025/09/economic-capital-a-new-measure-of-bank-solvency/}
}

You may also be interested in:

Bank Profits and Shareholder Payouts: The Repurchases Cycle

How Shadow Banking Reshapes the Optimal Mix of Regulation

A Check‑In on the Mortgage Market

Understanding the economic and financial consequences of natural disasters is a major concern for researchers and policymakers. The way in which overlapping natural disaster systems interact, as exemplified by the recent fires in Los Angeles being exacerbated by strong winds, is a major area of study in environmental science but has received comparatively little attention in the economics literature. Examining these potential interactions would likely be important for financial institutions, since such assessments would, in many instances, increase the estimated financial impact of a given natural disaster. In our recent Staff Report, we develop a method of identifying disaster systems in natural disaster data, such as the Spatial Hazard Events and Loss Database (SHELDUS), and use it to argue that the economics and finance literatures may have overlooked some sources of systemic risk.

What Is Clustering? 

We define clustering in natural disaster occurrences as the tendency for natural disasters to be concentrated in certain geographic regions and/or short periods of time. In this post, we focus on spatial clustering (that is, clustering in geographic regions), but the Staff Report also presents results on temporal clustering (that is, clustering across time). Clustering could have important implications for how we understand natural disasters if they have spatial spillover effects. For example, if neighboring counties are dependent on common emergency resources for aid in recovery, these resources may be strained if all of these counties are simultaneously affected by disasters. For these reasons, it is important to identify clusters of natural disasters across space in economic data on natural disasters. 

A Novel Approach for Identifying Natural Disaster “Clusters” 

For a given pair of neighboring counties, we ask whether they both experience damages from the same hazard type in the same month, as reported by SHELDUS. If so, they are treated as being part of the same cluster. This approach is repeated until no additional county pairs can be linked. In the left panel of the figure below, we see the cluster of counties that can be linked to Harris County, Texas (Houston) when Hurricane Harvey struck in August 2017. In comparison, the right-hand panel shows the footprint of counties identified in the official Hurricane Harvey Disaster Declaration by FEMA. The similarity of the identified cluster with FEMA’s declared disaster area shows that our algorithm can identify major disasters in the data with reasonable accuracy. 

Comparing the Harris County August 2017 Spatial Cluster (left) with the FEMA Hurricane Harvey disaster (right) 

Why Does “Clustering” Matter? 

Natural disaster damages data that are aggregated at the cluster level may have different distributional properties compared to standard panel data, which are measured at the county level. The chart below shows a comparison of the damages in the standard SHELDUS disasters data (that is, panel data at the county level), relative to a dataset of damages aggregated at the cluster level using our method. Interestingly, we tend to observe more extreme damages when analyzing damages at the cluster level. The difference between the distributions of damages using county- and cluster-level data highlights how it may be easier to capture the effects of extreme disasters when incorporating clustering into analyses of natural disaster outcomes. 

The Distribution of Disaster Damages According to Clusters and Counties 

To further explore whether larger clusters tend to be more damaging, we look at how disaster damages vary according to the size of a cluster, measured by the number of affected counties in the cluster. The chart below suggests that as relative size increases, average damages grow quickly. This chart, along with additional analysis in our Staff Report, suggests that counties tend to experience disproportionately more disaster damage when they are part of clusters that experience large amounts of damage.  This is suggestive of the existence of the types of disaster damage-related spillover effects that were discussed above.  

Differences in Disaster Damage in Terms of Cluster Size 

Lastly, we ask whether certain hazard types appear to cause different levels of damage depending on whether county- or cluster-level data are used. The chart below displays the relationship between average damages for various hazard types based on county-level data and those based on cluster-level aggregates. The scatter points lie above the 45-degree line, implying that all hazard types appear more destructive when using cluster-level aggregates rather than county-level data. This effect is especially pronounced for certain hazard types: Droughts are the ninth-most severe hazard type when using county-level data but are the second-most severe hazard type when aggregating damages to the cluster level, possibly because the average drought occurs in a cluster of about thirty counties, relative to an average cluster size of four counties across all hazard types. Damages from droughts therefore tend to be spread out across more counties. As a result, analyses of disaster damages at the county level may lead researchers to underestimate the severity of certain hazard types when those hazard types tend to occur in large clusters. 

Relationship Between Spatial Cluster Damage and County Damage by Hazard Type 

Final Words 

Inspired by an important idea from the environmental science literature, we develop a method for identifying clusters of disasters. We show that this approach is economically meaningful as it illustrates the heterogeneities in damages by natural disaster type. Failing to account for clustering may have implications for both policymakers and practitioners. For instance, if clustering is ignored, policymakers may insufficiently prepare for certain hazard types that tend to occur in large spatial clusters, such as droughts. Moreover, financial institutions may not correctly quantify natural disaster risk in their portfolios with respect to regions that are potentially exposed to low-probability, high-impact disasters. Finally, if disaster damages are correlated across different regions due to the phenomenon of spatial clustering, it may be difficult to obtain insurance for assets located in such areas. This could increase the likelihood of credit rationing in regions exposed to natural disasters, especially in markets where insurance is important, such as the real estate market. Therefore, our project sheds light on a potential source of systemic risk that banks, insurers, and policymakers may want to take into account. 

Jacob Kim-Sherman is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

Lee Seltzer is a financial research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

 
@article{Kim-ShermanSeltzer2025,
    author={Kim-Sherman, Jacob and Seltzer, Lee},
    title={What Is Natural Disaster Clustering—and Why Does It Matter for the Economy?},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={September 2},
    year={2025},
    url={https://libertystreeteconomics.newyorkfed.org/2025/09/what-is-natural-disaster-clustering-and-why-does-it-matter-for-the-economy/}
}

You may also be interested in:

Flood‑Prone Basement Housing in New York City and the Impact on Low‑ and Moderate‑Income Renters

Small Business Recovery after Natural Disasters in the Fed’s Second District

Potential Flood Map Inaccuracies in the Fed’s Second District

What is good management, and how is it transmitted across firms and plants? In a recent paper, we use survey and administrative data, coupled with a structural model of management, to explore these questions. We show that well-managed manufacturing firms—that is, firms that adopt more structured management practices described below—not only open and acquire more plants, but also close and sell more plants. Through this process, the firms transmit their management practices to new plants. These facts, taken together, imply that acquisitions can increase aggregate productivity by allowing well-managed firms to take over poorly managed plants and improve their management practices. 

Measuring Management 

We use two large-scale surveys to measure management practices: the U.S. Census Bureau’s Management and Organizational Practices Survey (MOPS), which covers manufacturing firms in the United States, and the World Management Survey (WMS), which covers manufacturing firms in thirty-eight additional countries. Both surveys pose a series of questions about three areas of management practices: monitoring, target setting, and people management.  

The monitoring section of the surveys asks firms about how they track information to monitor and improve their production process. For example, plant managers are asked how many performance indicators are monitored at each establishment, how they are displayed, and how often they are reviewed by plant managers and non-managerial workers. Firms that track multiple indicators, display them clearly, and review them frequently receive higher management scores. 

The target setting section of the surveys asks firms about how production targets are designed and how realistic they are. For example, plant managers are asked about the time frame and difficulty of meeting their production targets. Firms that have both short- and long-term production targets that are demanding but not impossible to achieve receive higher management scores. 

Finally, the people management section of the survey asks how performance is incentivized for managers and non-managers through promotion, bonuses, and dismissal policies. For example, plant managers are asked if promotions are based solely on performance and ability, or on other factors such as tenure or family connections. Firms that base promotion and compensation decisions on individual performance receive higher management scores.   

We use the answers to these surveys to create an overall management score for each plant and firm that ranges from zero to one, with higher scores indicating the adoption of more structured management practices. Using the MOPS in conjunction with other data from the U.S. Census, including the Census of Manufacturers and Longitudinal Business Database, we can track plant openings, closings, and acquisitions, enabling us to investigate the relationship between the management practices and dynamics of a given firm. 

Management and Firm Growth 

We find that well-managed firms in the MOPS exhibit higher plant turnover: they open, close, acquire, and sell more plants. 

We consider four margins of firm growth: entry (plants that are opened), exit (plants that are closed), acquisitions (plants that change ownership to the focal firm from another firm), and disposals (plants that change ownership from the focal firm to another firm). We calculate the entry rate as the number of plants that were opened divided by the initial number of plants managed by the firm, then measure these growth rates over five-year intervals. (We perform similar calculations for the other margins of growth.) 

The chart below shows that firms with higher management scores open more new plants. This comports with prior evidence that well-managed firms are generally more productive and grow faster. Interestingly, firms with higher management scores also close a larger share of their plants. While prior research suggests that well-managed plants are less likely to exit, these results are limited to “surviving” firms that are present throughout the five-year period. So, although individual well-managed plants are less likely to exit, well-managed firms still exit a larger share of plants. 

Well-Managed Firms Have Greater Entry and Exit Rates 

Echoing the results for entry and exit, the next chart extends this analysis to acquisition and disposal of plants on the M&A market. The data show that well-managed firms acquire more plants, and that they also dispose of more plants.  

Taken together, these findings indicate that firms with better management practices exhibit greater plant turnover. This trend parallels patterns observed in private equity takeovers, where high rates of plant churn are common. Notably, on both the entry/exit and acquisition/disposal margins, the net gains are positively correlated with management quality, so well-managed firms still grow more overall. 

Well-Managed Firms Have Greater Acquisition and Disposal Rates

Transmission of Management Practices 

Given the relationship between management quality and plant turnover, we next turn our attention to how management practices spread across plants within a firm. We are interested in understanding the extent to which management is non-rival. That is, can management practices be shared freely across plants within a firm as a type of technology, without diminishing their availability? Or, is management more akin to capital, with the presence of specific managers or equipment being necessary to spread management practices across plants?  

We find evidence that management is at least partially non-rival: good management practices spread to new entrant plants and to newly acquired plants. In the paper, we show that the management scores of new entrant plants are positively correlated with the management scores of their parent firms in the MOPS. We also show that acquired plants in the MOPS converge in management to their acquirer firm. Plants that are purchased by well-managed firms from poorly managed firms increase their management scores and revenue. 

We use WMS data to take a closer look at how management is transferred through acquisitions, running event study regressions to assess the effect of acquirer firm management on target firm performance. As shown in the chart below, we find that acquirer management scores have a positive effect on target revenue. Interestingly, the positive effect of acquirer management on revenue doesn’t show up until four years after acquisition, suggesting that it takes time for the purchased firm to implement the management practices of the acquiring firm. In the paper, we show that takeovers by well-managed firms also cause acquired firms to increase their productivity and decrease their employment and capital, suggesting that they can produce more efficiently post-acquisition. 

Performance Rises after Acquisition by Well-Managed Firms 

Aggregate Implications 

In the paper, we develop a model of firm dynamics to explore the implications of these relationships for aggregate outcomes. Using the model, we assess the effect of restrictive M&A policy and find that banning acquisitions has negative effects on productivity and output. When we turn off M&A, average management quality and average revenue both decrease by about 13 percent. This is because acquisitions reallocate plants from poorly managed to well-managed firms, which increases overall output and management quality. This highlights the important role of the M&A market in resource allocation. 

We also use the model to estimate the contribution of management to productivity gaps across countries, and find that, on average, management quality accounts for a 12 percent productivity gap with the U.S.—about one-fifth of the total average productivity gap between non-U.S. firms and U.S. firms. The bulk of this management gap is explained by absolute differences in management practices (that is, firms in other countries on average exhibit less structured management practices than the average practices in the United States), but one-quarter is explained by reallocation. That is, the covariance between management quality and firm size is larger in the United States, suggesting that U.S. firms are better able to allocate resources to well-managed firms. These findings are consistent with the idea that the allocative effects of management, realized through organic growth and acquisitions, have important consequences for aggregate output. 

Nicholas Bloom is the William Eberle Professor of Economics at Stanford University.

Jonathan Hartley is an economics Ph.D. Candidate at Stanford University.

Raffaella Sadun is the Charles E. Wilson Professor of Business Administration at Harvard Business School.

Rachel Schuh is a research economist in the Federal Reserve Bank of New York’s Research and Statistics Group. 

John Van Reenen is Ronald Coase School Professor at the London School of Economics and Digital Fellow, Initiative for the Digital Economy at the Massachusetts Institute for Technology.

 
@article{NicholasBloom,JonathanHartley,RaffaellaSadun,RachelSchuh,andJohnVanReenen2025,
    author={Nicholas Bloom, Jonathan Hartley, Raffaella Sadun, Rachel Schuh, and John Van Reenen},
    title={How Firms Spread Good Management},
    journal={Liberty Street Economics},
    note={Liberty Street Economics Blog},
    number={August 13},
    year={2025},
    url={https://libertystreeteconomics.newyorkfed.org/2025/08/how-firms-spread-good-management/}
}

You may also be interested in:

Are Businesses Absorbing the Tariffs or Passing Them On to Their Customers?

Nonbanks and Banks: Alone or Together?

The Origins of Market Power in DeFi