Federal Reserve Economic Data: Your trusted data source since 1991

The FRED® Blog

The St. Louis Fed’s Financial Stress Index, version 3.0

In 2010, the St. Louis Fed introduced its St. Louis Fed’s Financial Stress Index (STLFSI), which quantifies financial stress in the U.S. economy using 18 key indicators of financial market conditions—7 interest rates, 6 yield spreads, and 5 other indicators. This index, of course, can be found in FRED.

The STLFSI uses principal component analysis (PCA) to calculate the “factors” most responsible for the co-movement of several variables. By relying on multiple types of indicators, the STLFSI captures a broad, robust concept of overall financial stress. Just last year, we slightly revised the index’s methodology (creating the “STLFSI 2.0” or “STLFSI2”) to account for trends in several of the series. We’ll be revising the index again, and this post describes the motivations and details of this revision.

The London interbank offered rate, or LIBOR, measures the average interest rate at which major banks lend to each other short-term, unsecured (i.e., non-collateralized) loans. Lending to another private institution always has the risk that the institution will be unable to repay its loans, and the spread between the LIBOR and “riskless” interest rates over the same period helps quantify financial credit market risk. An increase in credit risk, all else equal, will increase the STLFSI.

Two of the indicators used in the STLFSI rely on the LIBOR: the yield difference (“spread”) between the 3-month LIBOR and the overnight index swap (the LIBOR-OIS spread) and the spread between the 3-month Treasury bill and the 3-month LIBOR (the TED spread).

But, starting this year, the LIBOR is being slowly discontinued, and Fed officials have encouraged the use of alternative measures in the meantime.* So, we are revising the STLFSI to account for this change.

Many rates have been suggested by regulators and market participants as a replacement. We, like many, have decided to replace LIBOR with the secured overnight financing rate (SOFR), which tracks the cost of short-term borrowing using transaction data on loans—collateralized by U.S. Treasury securities—in the overnight repo market. Proponents of the SOFR—including the Federal Reserve Bank of New York—argue it is a more accurate measure of bank borrowing costs than the LIBOR. The 90-day average SOFR also closely tracks the 3-month LIBOR.

One difference is that the LIBOR covers unsecured loans, while the SOFR covers secured loans (collateralized with Treasuries). Credit risk matters less in the latter case since the lender receives collateral if the borrower doesn’t pay back the loan. We see this in the graph above, where the SOFR tends to be lower than the LIBOR—reflecting the smaller risk of collateralized lending (and, thus, cost of borrowing). Nonetheless, its movements likely capture some information on changing credit risk since lenders prefer liquid cash over illiquid collateral—as evidenced by the SOFR’s co-movement with LIBOR.

A challenge in switching from LIBOR to SOFR is that the latter has a much smaller number of observations—it begins in 2018. We decided on a simple fix: We estimate what past SOFR spreads would have been, based on the LIBOR rate each day. We do this by calculating simple linear regressions that regress the SOFR spreads on their LIBOR counterparts, using average weekly observations from the SOFR’s introduction through the end of 2021, and use the regression’s estimates in our new STLFSI calculation for the years before the SOFR was introduced.

For the past several weeks, we have been tracking the new STLFSI (3.0) and comparing it with the STLFSI 2.0. As seen in the graph below, the correlation between STLFSI 2.0 and 3.0 is very high, about 0.99.

Still, there are some small but notable differences between the two indices. The biggest period of divergence is the first year or so after the SOFR was introduced (2018-19)—which makes sense, since (as we saw in our first graph) the SOFR initially did not track the LIBOR as closely as it has more recently. More interestingly, the STLFSI 2.0 tended to be slightly higher than the STLFSI 3.0 during the Great Recession, whereas the STLFSI 3.0 has tended to be higher than the STLFSI 2.0 during the COVID pandemic; indeed, it has been consistently about 0.05 index points higher than the STLFSI2 in the last year. Despite these differences, the two indices nonetheless provide consistent signals of above- or below-average financial market stress, with few occasions where one is positive and the other negative. Thus, we are confident that the new STLFSI will continue to serve as a reliable indicator for monitoring financial conditions.

* Former Federal Reserve Governor Randall Quarles noted in a speech last year that the LIBOR would not be available for any new contracts beginning in 2022. Governor Quarles also said that the Fed and other regulators sent a letter to banking organizations they oversee stating that “after 2021, the use of LIBOR in new transactions would pose safety and soundness risks.” These supervised institutions were “encouraged” to seek out an alternative reference rate for new contracts beginning on January 1, 2022. As we discussed above, the recommended alternative reference rate is the SOFR.

Why are the Fed and other regulatory institutions urging financial institutions to discontinue the use of LIBOR? As Governor Quarles and others noted, years after the STLFSI’s release, regulators have highlighted LIBOR’s shortcomings over several years. Quarles stated:

The principal problem with LIBOR is that it was not what it purported to be. It claimed to be a measure of the cost of bank funding in the London money markets, but over time it became more of an arbitrary and sometimes self-interested announcement of what banks simply wished to charge for funds.

How these graphs were created: For the first graph, just search for the St. Louis Financial Stress Index and select the series that is not discontinued. For the second graph, search for “90-day SOFR”: From the “Edit Graph” panel, use the “Add Line” tab to and search for and select “3-month LIBOR.” For the third graph, take the first and add a line searching for “STLFSI2.”

Suggested by Aaron J. Amburgey, Kevin L. Kliesen, Michael W. McCracken, and Devin Werner.

Are we still in a recession?

What to expect from the NBER business cycle dating committee

The Downturn and Rebound

  • April 29, 2020: In its advance estimate, the Bureau of Economic Analysis (BEA) reported that real GDP for the first quarter of 2020 fell at a 4.8% annual rate.
  • May 8, 2020: The Bureau of Labor Statistics reported that nonfarm payrolls fell by 20.5 million in April—the largest one-month percentage decline on record (dating back to 1939).
  • June 8, 2020: The National Bureau of Economic Research Business Cycle Dating Committee (NBER BCDC) announced that the 128-month expansion (the longest in U.S. economic history, dating back to 1854) ended sometime in February 2020.
  • Since then, the U.S. economy has rebounded sharply, posting large increases in real GDP and nonfarm payroll employment and a large decline in the unemployment rate. But the NBER BCDC hasn’t yet announced an end to the recession…

The BCDC’s Methods

The BCDC patiently assesses business cycle peaks and troughs. For example, they announced that the trough of the 2007-2009 recession occurred in June 2009 only on September 20, 2010—which is a lag of 15 months.

The BCDC also emphasizes economywide economic indicators. In their view, dating peaks and troughs is best accomplished by looking at measures of activity that cut across all sectors of the economy, rather than a small number of key sectors (such as the Federal Reserve’s industrial production index, which measures output produced by the nation’s manufacturers, utilities, and mining industry). In their June 8 announcement, the BCDC indicated that real GDP and real gross domestic income (GDI) are the two “most reliable” comprehensive measures of economic activity.

The FRED graph above shows real GDP data from the BEA: Real GDP fell in the first and second quarters of 2020, but then rebounded in the third quarter. However, unlike real GDP, real GDI isn’t yet available for the third quarter because the BEA hasn’t yet reported corporate profits—a key component of GDI. Corporate profits will be reported in the second estimate of GDP, scheduled for release on November 25.

What Else Do They Look At?

Over time, the BCDC has examined several comprehensive monthly indicators, such as real manufacturing and trade sales, nonfarm payroll employment, and civilian employment. In its September 2020 announcement, the BCDC emphasized that real personal consumption expenditures and real personal income excluding current transfer payments are the two broadest measures of aggregate expenditures and aggregate income. Use this FRED dashboard to follow such comprehensive indicators. April 2020 was the trough month of all five of these indicators. Moreover, each of the indicators has since risen sharply, consistent with the increase in real GDP.

What’s Different This Time?

As noted above, the BCDC generally prefers to wait until there’s conclusive evidence that the economy has transitioned from a period of recovery to expansion. The unique features of this pandemic-spawned recession have, as the BCDC noted on June 8, 2020, “resulted in a downturn with different characteristics and dynamics than prior recessions.” So, while the data suggest that an economywide trough in economic activity occurred sometime in the spring, the pandemic remains a key driver of economic policy and the behavior of many governments and individuals worldwide. From that standpoint, the length and strength of the recovery is uncertain.

How this graph was created: Search FRED for real GDP, select the series, and start the sample period on 2014-01-01.

Suggested by Kevin Kliesen.

View on FRED, series used in this post: GDPC1

The decline in industrial production: One for the ages

On Tuesday, April 15, the Federal Reserve released the industrial production (IP) index for March. You have to go to the very far right data point in the FRED graph to see it, but industrial activity plunged in March because of the economic effects stemming from social-distancing orders in response to the COVID-19 pandemic. Millions of businesses have closed or been disrupted, and mass layoffs have occurred. But the March IP index of 103.66 is still far higher than the level registered during the depth of the recession and financial crisis, which was 87.07 in June 2009.

The IP index is one of the nation’s longest continuously produced economic indicators, starting in January 1919. It measures production (real output) of manufacturers, mining (e.g., oil and natural gas), and electric and gas utilities and steadily increases over time; but it is highly sensitive to the state of the economy and falls during recessions, generally proportionate to the depth and duration of the recession. The 2007-2009 recession and financial crisis is a prime example.

FRED can help us compare this recent decline in IP against the entire history of the series. And the next graph shows one way to do this: month-to-month percentage changes. Measured from its February level, industrial activity fell 5.4% in March. This percentage decline is the largest in a long time, since January 1946 (-5.6%), when U.S. factories transitioned from producing primarily wartime goods to producing civilian goods for a peacetime economy. And the largest percentage decline in the series was -10.4%, in August 1945.

In fact, as this graph shows, the retooling of the U.S. economy in 1945 produced larger monthly percentage declines in IP than those during the Great Depression and the deep 1937-38 recession. So, March’s COVID-19-driven plunge in activity, while historically large, falls far short of previous declines in activity. End of story? Not quite.

Another way to gauge the historical magnitude of the March decline in IP is to benchmark it against the historical standard deviation of monthly percent changes. The standard deviation is a statistical measure of changes, or dispersion, relative to the mean (average) of the series. Most of the time, monthly percentage changes are within plus or minus one standard deviation. At times, though, large changes are well outside the bounds of one standard deviation. The larger the percentage change outside the series’ standard deviation, the larger its historical significance.

Now, the second graph also shows that the month-to-month percentage changes have become smaller over time. For example, compare the period before and after 1947—effectively, the transition to a post-WWII economy and the post-WWII economy itself. We can see that volatility—the swings between peaks and troughs—was much larger in the earlier period than in the later period. But to verify this, we’ll need to look at some statistics.

The table below shows the largest percentage declines in IP and their respective sample standard deviation over two intervals: February 1919 to December 1946 and January 1947 to March 2020. The standard deviation of monthly percent changes in IP was 3.29% in the first period and 0.96% in the second period. Hence, the standard deviation in the first period was three times as large as the second period. The table’s right-most column shows the ratio of the two statistics. By this metric, the March 2020 decline in industrial production was the biggest decline on record relative to its standard deviation.

Thus, in this sense, the March decline in IP was one for the ages.

 

Statistics on Monthly Percentage Changes in IP

  Minimum Standard Deviation Minimum/Stnd. Dev.
1919 to 1946 -10.38 3.29 -3.16
1947 to Present -5.40 0.96 -5.61

 

How these graphs were created: For the first graph, search for “Industrial Production” and it should be your first choice. For the second graph, start with the first and use the “Edit Graph” panel to change units to “Percent Change.”

Suggested by Kevin Kliesen.

View on FRED, series used in this post: INDPRO


Subscribe to the FRED newsletter


Follow us

Back to Top