key: cord-0685226-cevrq9rz authors: Bordo, Michael D.; Duca, John V. title: How New Fed Corporate Bond Programs Cushioned the Covid-19 Recession date: 2022-01-21 journal: J Bank Financ DOI: 10.1016/j.jbankfin.2022.106413 sha: 1338594d4779522cdcbc1b889a98316837b242b1 doc_id: 685226 cord_uid: cevrq9rz In the financial crisis and recession induced by the Covid-19 pandemic, many investment-grade firms became unable to borrow from securities markets. In response, the Fed not only reopened its commercial paper funding facility but also announced it would purchase newly issued and seasoned corporate bonds rated as investment grade before the Covid pandemic. We assess the effectiveness of this program using long sample periods, spanning the Great Depression through the Great and Covid Recessions. Findings indicate that the announcement of corporate bond backstop facilities helped stop risk premia from rising further than they had by late-March 2020. In doing so, these backstop facilities limited the role of external finance premia in amplifying the macroeconomic impact of the Covid pandemic. Nevertheless, the corporate bond programs blend the roles of the Federal Reserve in conducting monetary policy via its balance sheet, acting as a lender of last resort, and pursuing credit policies. In the financial crisis and recession induced by the Covid-19 pandemic, the ability of firms and municipal governments to borrow from securities markets dried up, as did access by small and mid-sized firms to bank loans. The massive credit squeeze risked seriously damaging the real economy via a financial accelerator mechanism. This credit crunch was manifest in a spike in the corporate Baa-10-year Treasury bond yield spread, a long-standing measure of systemic risk (see Figure 1 ). In response, the Fed not only reemployed all the new tools it created during the Great Recession, but also greatly expanded its credit-easing and lender of last resort role in several ways. This study focuses on the Fed's announcement that it would buy exchange traded funds (ETFs) of seasoned bonds and newly issued bonds of corporations rated as investment grade before the Covid pandemic. It did so with the explicit backing of the Treasury, which covered any default losses from such purchases. This novel policy was not used in 2008 since without Treasury indemnification against losses, the Fed believed that the riskiness of corporate bonds implied that it did not have the requisite authority. In addition, the Fed in 2008 did not foresee the magnitude and impact of the severe widening of the corporate-Treasury bond spread, whose peak reached highs not seen since 1935 (Duca 2017). 1 As Figure 1 illustrates, the spread between yields on Baa-rated corporate and 10-year Treasury bonds (BaaTr) in the recent crisis rose in line with the weekly insured unemployment rate until the Fed announced its corporate bond program in March 2020. A careful splicing of different unemployment rate series enables us to assess the effectiveness of recent Fed corporate bond interventions over a long sample that spans the Great Depression through the Great and Covid-19 Recessions. Findings indicate that the announcements of corporate bond backstop facilities prevented corporate bond risk premiums from rising further above pre-GFC averages than they did by late-March 2020. By doing so, these programs limited the role of external finance premia in amplifying the macroeconomic impact of the pandemic and the risk that a panic-induced wave of corporate bankruptcies could worsen the downturn and prolong the recovery from it. Thus, the corporate bond facilities can be seen as a new means by which the Fed pursues its full employment and price stability objectives rather than as an expansion of its mandate. As shown in Figure 2 , a motivation for this new tool is the rise of corporate bonds and the fall of depository loans as sources of external debt for nonfinancial corporations. Our study is organized as follows. Section 2 discusses the evolution of the Fed's response to cushion the economy from financial shocks. Section 3 provides details on the Fed's corporate bond facilities and Section 4 reviews how our paper contributes to the emerging literature on them. Section 5 presents our models of corporate bond premia and Section 6, our data and variables, some of which are novel and extend to 1929. Section 7 presents regression and simulation results. Section 8 discusses how the corporate bond programs differed from other policy actions and presents some robustness checks. The conclusion provides a broader perspective on our findings. The Baa-Treasury spread has long been used to analyze the macro-economy and as a gauge of financial crises dating back to at least Friedman and Schwartz (1963) . Since Bernanke (1983) , the spread has been made more popular by the credit and financial frictions literature, including Gilchrist (1996, 1999) , Bordo (2008) , Mishkin (1991) Mishkin and White (2002) . 2 A major reason is that this spread correlates well with banking panics, severe financial crises, extreme political shocks and other events as shown by Bordo (2008) . The Federal Reserve was founded to provide financial stability by acting as a lender of last resort to prevent the banking panics that plagued the national banking era (1865 to 1914). The Fed was mandated to follow well-established central banking practice (Bordo and Wheelock, 2011) and to lend freely on the basis of sound collateral (eligible banker's acceptances or commercial paper). A rule of thumb for central banks' lender of last resort policy, based on British precedent, was Bagehot's rule (1873) which stated that in the face of a banking panic a central bank was to lend freely but at a penalty rate to solvent but illiquid financial institutions. Bagehot's rule also has been interpreted as lending freely to the money market and not to individual banks (Bordo 2014 ). The Fed successfully prevented a banking panic in 1920 Metrick 2013, Tallman and White 2020) and the New York Fed successfully provided liquidity to the New York money center banks during the October 1929 Wall Street Crash to prevent a panic (Friedman and Schwartz 1963) . However, the Fed did not prevent banking panics between 1930 and 1933 which Friedman and Schwartz (1963) argued induced a one-third decline in the money supply that made the Great Contraction 1929-1933 so severe. Bernanke (1983) supported their ‗money hypothesis' but argued that the banking failures propagated the depression by raising the cost of credit intermediation by severing the link between saving and investment that banks provided. Bernanke‗s focus on credit led to the concept of the ‗financial accelerator‖ Gilchrist, 1996 and 1999) . A large financial shock leading to a collapse of asset prices reduces the net worth of firms and households, which reduces the collateral available for loans. Commercial banks cut lending which reduces household consumption and business investment. This lowers real output and prices, which in turn reduce net worth and bank lending. This process leads to debt defaults and bankruptcies, spurring bank failures and further credit impairment. In addition, deflation leads to Irving Fisher's (1933) debt deflation, which further reduces net worth, amplifying the downward spiral. Tightening credit conditions also affect nonbank financial intermediaries and financial markets leading to a general collapse of credit. A key indicator of credit turmoil is the Baa-10-year Treasury bond spread which tracks not only the effects of a credit crunch leading to defaults, but also a shortage of liquidity as occurs in banking panics. After the Great Contraction, the banking panic problem was greatly reduced by important reforms to the Federal Reserve and the banking system including the creation of the Federal Deposit Insurance Corporation (FDIC) and adding Section 13.3 to the Federal Reserve Act greatly. To overcome the severe restrictions on the Fed's discount window lending during the Great Contraction (Bordo and Wheelock 2011) , section 13.3 allowed the Federal Reserve to lend to non-member banks and other institutions on sound collateral in -unusual and exigent circumstances.‖ Although there were several banking crises from the 1970s to the Global Financial Crisis (GFC) these were not classic liquidity driven panics but rather solvency crises (Bordo 2014) . They were addressed by fiscal bailouts and not by lender of last resort actions (Bordo and Meissner, 2016) . The GFC, which began with a collapse in house prices and was centered in the shadow banking system, involved a massive credit crunch amplified by fears of counterparties holdings of toxic financial derivatives based on mortgages of varying quality. This fear was manifest in a spike in the Baa-Treasury spread as well as other spreads, such as the TED spread (Libor-Treasury bill), the commercial paper -Treasury bill rate spreads, and the gap between the 30year mortgage and 10-year Treasury interest rates. The Fed initially viewed it as a liquidity panic and greatly expanded its discount window facilities to give many financial institutions access to the discount window. The credit crunch spread through the plumbing of the financial system to the repo market a critical source of bank funding. This led the Fed to create facilities to unclog the arteries of the financial system. Via the Commercial Paper Funding Facility, the Fed bought newly issued, topgrade paper and helped cap the paper-bill spread (Duca, 2013a) . The Fed also bought Aaa-rated debt of lenders via the Term Asset-Backed Securities Loan Facility, which helped keep lender funding costs and hence loan interest rates from soaring (Agarwal, et al. 2010) . Moreover, the tightening of credit set in motion the financial accelerator mechanism that Bernanke (1983 Bernanke ( , 1995 outlined for the Great Contraction and for which Bernanke and Gertler (1990) provided theoretical underpinnings. Such considerations led the Fed to develop radical new facilities and buy large quantities of mortgage-backed securities (MBS) to keep credit flowing through the system. It later became apparent that more important than a liquidity shortage was the potential insolvency of the investment and universal banks that held the toxic derivatives in off-balance sheet special investment vehicles (SIVs). This problem was finally solved by recapitalizing banks with TARP funds following a series of stress tests, both of which allayed counterparty risk. The March 2020 Covid-19 crisis had elements of both a liquidity crunch and a credit crunch reminiscent of the GFC, as households cut work hours and consumption and as firms cut investment and payrolls with negative effects magnified by government-mandated lockdowns. The panic was reflected in a spike in the Baa-Treasury bond spread. It became quickly apparent to the Fed that massive liquidity injections were needed to prevent not only widespread defaults, but also a collapse of credit networks and financial accelerator effects that would amplify the downturn. Quite notably the Baa-Treasury spreads stopped rising on March 23, 2020, when the Fed announced that it would set up the PMCFF and SMCFF. Furthermore, the subsequent purchases by the Fed were under 50 billion by the early July 2020-far below the limits on the size of the facilities (see Table 1 )-with the vast bulk being purchases of ETFs by the SMCFF. Instead of reflecting a balance sheet effect (as with QE), this pattern reflects a strong -backstop‖ effect from announcing the facility by a central bank having a great ability to expand its balance sheet, with an announcement effect evident seven weeks ahead of the Fed's initial corporate bond purchase. The Fed's corporate bond interventions have spurred new research that generally takes one of two approaches. The first assesses how a wave of redemptions by household investors triggered large fire sales of corporate and municipal bonds by mutual funds that fueled a liquidity crisis in the corporate bond (see Falato, Goldstein, and Hortaçsu, 2020) and Haddad, Moreira, and Muir, 2020) and municipal bond (Li, O'Hara and Zhou, forthcoming) markets. Falato, et al. (2020) show that bond mutual outflows were greater for funds that invested in more Covid sensitive firms, and that the Fed announcements of the creation and expansion of the PMCCF and SMCCF spurred larger reversals of flows back into funds more exposed to bonds that were eligible for purchase by the new Fed facilities. Haddad, et al. (2020) find that the announcements of the creation of the two corporate bond programs disproportionately reversed how much corporate-Treasury bond spreads had risen relative to expected default risk (as tracked by CDS spreads) in the early spring. In particular, the reversals were greater for bonds that were eligible for purchase by the facilities. The second strand of literature tests whether the transactions costs or yields on bonds that were eligible for purchase fell more than those for ineligible bonds after the program announcements. O' Hara and Zhou (2021) show that trading costs for securities dealers rose and dealer inventories of corporate bonds fell early in the pandemic. Using a diff-in-diff approach, they find that dealer transactions costs fell for newly issued bonds eligible for the PMCCF versus those that were ineligible. Analogously, they find that transactions costs for previously issued bonds fell for those that were eligible for purchase by the SMCCF relative to those that were ineligible. Three other papers apply a diff-in-diff or regression discontinuity approach to bond yields and spreads. Boyarchenko, Kover, and Shachar (2020) , D'Amico, Kurakula, and Lee (2020), and Gilchrist, et al. (2020) identify how the Fed's announcements of bond facilities for investmentgrade debt and the extension of eligibility to some downgraded -fallen angel‖ firms lowered bond yield spreads and returns more for eligible versus ineligible bonds. Boyarchenko, et al. (2020) find that improved functioning of secondary markets lowered benchmark yields off which primary securities are priced, thereby lowering yields on new issued, eligible bonds and by making dealers more willing to place issues. D 'Amico, et al. (2020) and Gilchrist, et. al (2020) find larger effects for debt with maturities closer to those bought by the two bond facilities than for longer maturities. 4 Another relevant finding is that much of the rise in yields on investment grade corporate bonds during the spring of 2020 owed not to rises in expected default risk, but rather to greater risk aversion or liquidity premiums. Haddad, et al. (2020) stress how corporate-Treasury bond spreads rose relative to CDS spreads, while Gilchrist, et al. (2020) analyze corporate bond yields controlling for default risk among other variables. Both studies find that these widenings reversed more for bonds eligible for purchase by the facilities when the facilities were announced. Together, these studies provide evidence of a notable backstop effect of the new Fed corporate bond programs. Nevertheless, these studies may understate the impact of the programs by focusing on differential effects on eligible versus ineligible bonds, thereby missing some of the indirect effects on yields on ineligible bonds, and also how the worsening economic conditions would have likely raised risk premia even higher absent the Fed intervention. In addition, data limitations prevent these studies from drawing information from prior business and financial cycles. Because background conditions can shift over time, cross-section results from one episode may not fully apply to other episodes. Our study complements these papers by analyzing bond spreads across several business cycles, which can help guide counterfactual estimates of how high the spreads would have risen absent the new facilities. Although our time series findings are consistent with many of these micro/cross section studies, our focus is not on establishing causality from the programs to bond yields, but rather on gauging their effects on limiting the amplification of the recession by limiting the nascent surge in spreads that was underway in early March 2020. The main corporate spread investigated is the gap between yields on Baa-rated corporate bonds (Moody's) and on long-term U.S. government bonds. Baa and Aaa corporate yields are available monthly since 1919. Of these, the Baa is more relevant because few firms are rated Aaa (only two in 2020) and because the impact of financial frictions on firm activity tends to be greater for firms that are less resilient than Aaa-rated firms. The 10-year Treasury yield series follows Duca (2017) in splicing three series on long-term Treasury yields. This spread (BaaTr) widens in recessions, especially in the Great Depression, Great Recession, and the Covid Recession ( Figure 3 ). The Baa spread also tends to be coincident with the business cycle (Duca, 1999) , in line with evidence that a consistently measured monthly unemployment rate (UR, Section 6.1) has moved with the BaaTR spread since the late 1920s. Before the Fed announced the corporate bond facilities, the Baa-Treasury spread mainly reflected not only measurable cyclical and secular risks for corporate debt and shifts in monetary policy, but also of hard-to-forecast shocks to liquidity, default risk, and risk aversion. Time series measures for each factor are limited by data availability and endogeneity. 5 Tracking the cyclicality of spreads over long periods requires consistent measures of the depth of cycles for which the unemployment rate, output gap, and capacity utilization are available monthly, and the insured unemployment rate, weekly. The correlation of the BaaTR spread is stronger with the square of the unemployment rate than with the level, while the spread's correlation with the output gap and capacity utilization is stronger in their levels. 6 While monthly results are similar using the latter two variables, we mainly proxy cyclical risks with the square of the unemployment rate (UR 2 ), which is expected to be positively correlated with BaaTR and is available monthly and weekly. That correlation, however, is affected by a major shift following the failure of Penn Central in April 1970 (PennCentral =1 since then, 0 before), which Jaffee (1975) also found 5 Fed interventions prevent using the commercial paper bill spread (Friedman and Kuttner) after 2007 or corporate spreads in modeling commercial paper (Duca, 2013) or municipal-Treasury spreads after February 2020. Endogeneity prevent using the yield curve as long-term Treasury yields affect its slope and the base off which corporate and muni spreads are measured. Also, QE affects the consistency of the yield curve over time. The post-2007 Fed balance sheet also limits using the TED spread as high excess reserves the spread's ability to track general market risk premia. 6 Swings in output around trend tend to be larger than rises in the unemployment rate, as noted Okun, 1970) who argued that labor market frictions gave firms incentives to adjust employment less than output. This is consistent with our findings that cyclical shifts in corporate risk premia are more correlated with the square of the unemployment rate rather than its level, while levels of the output gap and capacity utilization outperform their quadratic counterparts. The second major secular shift in BaaTR is an upshift in December 2000, when Congress approved the Commodity Futures Modernization Act (CFMA). As Bolton and Oehmke (2015) and Stout (2011) show, the CFMA exempts derivatives counterparties from the automatic stay in bankruptcy, enabling immediate collection from a defaulted counterparty, giving them a senior claim over most other bankruptcy claimants. The passage of CFMA was quickly followed by a surge in credit default swaps (Duca and Ling, 2020) . By giving bankruptcy priority to derivatives (mainly credit default swaps), the CFMA lowered the priority of bond investors and made bonds riskier (see Bolton and Oehmke 2015) . We find that a shift dummy (CFMA =1 since December 2000, = 0 otherwise) captures an upward level shift in the corporate-Treasury yield spread. 7 These considerations imply the following specification for the equilibrium spread: where each coefficient α i has an expected positive sign. According to stationarity tests (Appendix A), the levels of BaaTR and UR are not stationary at monthly and weekly frequencies, while their first differences are stationary. Owing to their construction, the same is the case for the two shift variables, PreAccord and CFMA. These tests indicate that the variables in eq. (1) have unit roots. Accordingly, we estimate eq. (1) monthly over 1929 to 2021 using a cointegration approach (Johanssen's (1995) ) to jointly estimate the long-run level and the short-run change. The latter first difference equation includes an error-correction term (the t-1 gap between the actual and equilibrium spread), lags of first differences of all the variables in the long-run equilibrium relationship and a vector X of exogenous event shocks. The inclusion of the latter is needed to yield uncorrelated residuals in the first difference model and has little effect on the estimated coefficients of the long-run (equilibrium) relationship. Building off eq. (1), we thus estimate: where EC t-1 ≡ BaaTR t-1 -BaaTR * t-1 , lag lengths minimize the SIC, Ω is a vector of coefficients for the X vector and the estimation allows for time trends in each variable in the long-run equation but not a trend in the long-run relationship. The coefficient β 1 on the error-correction term should be negative so that the spread converges toward its equilibrium), with the absolute magnitude implying the speed of error-correction. Note that the model implicitly imposes an almost instantaneous reaction of the corporate bond yield to the long-term Treasury yield, Given the short post-Covid sample, using higher frequency weekly data has the advantage of helping identify the effects of the pandemic and Fed interventions on the corporate bond market, but at the disadvantage of not spanning pre-1971 data. The weekly models have the advantages of using unemployment data that have not been spliced and avoid the need to control for regime shifts other than CFMA. Accordingly, the weekly model for the post-1970 sample simplifies to: This section provides details on the unemployment rate variables, how some models adjustments for Fed corporate bond programs, and on exogenous shock variables. The main cyclical variables are variants of the monthly civilian unemployment rate and the weekly insured unemployment rate from the initial claims for unemployment report. Monthly unemployment rates are based on data from the Bureau of Labor Statistics (BLS) that the BLS derived from the monthly household survey since 1948 and from other, earlier Census surveys. Readings from 1933 to 1943 are adjusted for time-varying employment in Depression-era federal job-creation programs that were excluded in calculating the unemployment rate. Adjustments use estimates from Darby (1976), who showed that official statistics from this period are not comparable with those from other periods, and notably overstate unemployment in the mid to late-1930s. As discussed in an appendix in Bordo and Duca (2020) , monthly readings on unemployment are adjusted based on Darby's estimates and for several breaks in data before 1948. Monthly data from the household survey spanning January 1948-present are spliced with 1940-47 data from an earlier Census survey and with 1929-39 data from the Conference Board. In weekly models our cyclical variable is based on the weekly, insured unemployment rate from the weekly initial claims report, available since 1971. The raw series has not consistently moved with the monthly unemployment rate (Cleary, Kwok, and Valletta, 2009) because the taxation of, and eligibility for, UI benefits have changed, as have the regional and unionized composition of employment which affect the proclivity of the unemployed to file for benefits and count as unemployed. To address this we multiply the seasonally adjusted weekly series by the centered, 12-week moving average of the ratio of the weekly unemployment rate to the monthly unemployment rate. 9 While the adjusted and unadjusted weekly unemployment rates are each significantly related to (cointegrated with) the Baa-Treasury spread and the spread significantly error-corrects toward its equilibrium, the adjusted series yields short-run residuals that are not significantly correlated in contrast to residuals from a model using the unadjusted series. The Fed's corporate bond programs were introduced to stem further rises in corporate bond spreads and their announcement on March 23, 2020 rapidly affected expectations and spreads well before the first bond purchases were made in May 2020. Essentially, the Fed's announced intervention ended further feedback from a worsening economy and investor concerns on corporate bond premiums. We track the former amplification effects-and to some extent other factors affecting risk premia around downturns-using the square of the unemployment rate. 10 To gauge the impact of the corporate bond program, we create a squared unemployment rate series that equals the squared unemployment rate series (UR 2 ) before the program 9 The smoothing of the adjustment parameter limits noise in the weekly, and to a lesser extent in the monthly series. 10 Other risk premia shocks are tracked by the lagged first differences in the risk premia and long-term variables. announcements and after the programs ended in December 2020. From the announcement to December 2020, the program-adjusted unemployment rate series (URCCF) is the maximum of the unemployment rate in late March (6.4 percent, the average of the last two weekly adjusted insured unemployment rates), and the weekly insured unemployment rate. This adjustment implies that the program had very large effects on the spread in spring 2020 and large effects in summer 2020, and then faded away by mid-November 2020 and December 2020 on a weekly and monthly basis, respectively. The adjusted unemployment rate series mirrors the capping of spreads by the bond backstop by reflecting that the bond programs had a larger effect at the height of the pandemic and that the backstop effects ebbed in the summer and fall of 2020, before disappearing before the program had ended. In other models, we made analogous adjustments based on the May 12 th start of Fed purchases of corporate bond ETFs but the announcement-based models fit notably better. Five monthly shocks are added to the model. Four occur in the Great Depression and three of these reflect temporary, large effects of bank failures and crises on the Baa spread in an era before deposit insurance and aggressive Fed lender of last resort policies. Models of the corporate Baa-Treasury spread are estimated using monthly data from 1929 to 2021 and weekly data over 1971 -2020. The advantage of the long sample is that it spans the Great Depression. The shorter sample, weekly models have more observations to assess the impact the Fed corporate bond programs. In our models, the unemployment rate terms are adjusted during the corporate bond programs to reflect that the announced programs essentially prevented the Baa spread from widening further as the economy deteriorated during the early onset of the pandemic. Table 2 reports results from monthly models that include the PennCentral shift in the cyclicality of the spread and the Treasury-Fed Accord and CFMA shift level shift terms in the long-run model, and the set of shock terms in the X vectors described in Section 3. In each model in Table 2 In the case of the models that ignore the Fed corporate bond facilities, this is reflected in Figure 4 which plots the equilibrium relationship from the full sample estimates from Model 6 and a counterfactual which replaces the squared unemployment rate with the squared unemployment rate that is adjusted for the corporate credit facilities (CCF) effect. Before the Covid pandemic hit the U.S., the estimated equilibrium generally trends with the actual spread, with upward spikes during periods of heightened risk that are hard to predict or forecast. Notice how much the equilibrium spread from the model that ignores the CCF effects exceeds the actual spread and the counterfactual equilibrium from the model using the CCF-adjusted unemployment rate (URCCF). Results are similar if the latter is replaced by using coefficients from the pre-Covid model having short-run controls (Model 2) and using the non-CCF adjusted unemployment rate. This is done in Figure 5 which focuses on a post-1999 sample. The model estimated over the pre-Covid era implies a peak of 6.0 percent in the equilibrium spread, which exceeds the actual April 2020 peak by 2.6 percentage points and is 3.5 percentage points above the peak in the equilibrium implied by the model accounting for CCF effects. The gap between the two equilibriums from The large disparities between the pre-Covid relationships and spreads in the Covid era imply that the error correction terms from models ignoring the Fed interventions overstate the extent to which the actual spreads exceed the equilibriums. As a result, and because the errorcorrection terms have negative signs, the short-run predicted changes from such miss-specified models exceed the actual changes during the pandemic. This will induce a string of negative model residuals and could result in significant serial correlation, which occurs in Models 3 and 5 that ignore the Fed programs. In contrast, the post-Covid models that account for how the Fed programs capped the amplification effects of unemployment (Models 4 and 6) have well behaved residuals, as is the case for the pre-Covid model that includes exogenous shocks (Model 2). It is important to note how much short-run exogenous shocks account for variation in the short-run changes in the Bas-Treasury spread in Table 2 . As expected, the dummies for the failure of the Bank of U.S., the Fed's short-lived QE experiment in 1932, and Lehman's demise were positive and significant, while the December 1933 change in the gold standard had a significantly negative effect. Of the models having short-run shock terms (Models 2, 5, and 6), the corrected R 2' s for the short-run model portions range between 0.44 and 0.46, which are reasonable for models of the first-difference of a spread. Moreover, these corrected R 2 's are 0.33 to 0.34 higher than those from corresponding models without them (Models 2 vs. 1; Models 5 vs. 3, and Models 6 vs. 4). In addition, the error-correction terms are significant in every model with the expected negative sign and imply sensible speeds at which the Baa-Treasury spread adjusts to its long-run equilibrium. Nevertheless, the annualized speeds of adjustment are also much higher in corresponding models that include the shock terms. For the pre-Covid sample, the estimated annual speeds of correction are 15 percentage points faster in the model with the controls (Model 2) than without them (Model 1). For full sample models that do not control for the impact of the corporate bond programs on unemployment effects, annualized speeds of adjustment are 11 percentage points faster in the model with the controls (Model 5) than without them (Model 3). Finally, for models which account for the new Fed programs, the annual speeds of adjustment are 15 percentage points higher with the controls (Model 6) than without them (Model 4). Results from estimating weekly models with post-1970 data are reported in Table 3 . Owing to their later sample start, these models omit the short-run Great Depression shock variables and the long-run Treasury-Fed Accord, and PennCentral variables. Models 1 and 2 exclude and include a set of short-run shock variables over a pre-Covid sample, respectively. The Paralleling the monthly models in Table 2 , the odd-numbered models in Table 3 omit the shock terms with only Models 4 and 6 adjusting the unemployment rate for the Fed's announced creation of the corporate bond programs. Similarly, Models 5 and 6 in Table 3 include the Covid shock terms with only Model 6 using the cap-adjusted unemployment rate. The Schwartz Information Criterion selected lag lengths on lagged first differences in the short-run model of 12 for the weekly models. As with the monthly models, we identify a significant and unique long-run cointegrating relationship in each weekly model, long-run coefficients are significant with the expected signs, and the corresponding short-run models have uncorrelated residuals in the pre-Covid sample. In pre-Covid sample models with short-run controls (Model 2), the long-run unemployment and CFMA coefficients and speeds of error-correction are significant and are not statistically different between the weekly and monthly models (the latter add in the Penn-Central shift effects): As with results using monthly data back to 1929, in the post-1970 weekly models extended into the Covid era, there is evidence that not controlling for the Fed's new corporate bond programs results in misspecification. In particular, the residuals from such models (Models 3 and 5 in Table 3 ) exhibit significant serial correlation according to VECLM statistics, whereas residuals are free from serial correlation in the models that account for the credit programs (Models 4 and 6). Indeed, the models that do not account for how the corporate bond programs attenuated the unemployment-amplifier effect, also suffer from long strings of predictions that overstate short-run changes in the spread (i.e., correlated negative residuals) after the Fed announced the creation of the facilities. Other short-run results are worth noting. First, the error-correction terms are negative and significant and imply sensible speeds at which the Baa-Treasury spread adjusts to its longrun equilibrium. In particular, the estimated coefficient implies that about 60 percent of the gap is corrected within a year. As expected for the weekly short-run shock terms, the estimated coefficients on DLehman, StockCrash87, and D911 are significant and positive while that for USDGrade is negative and significant. Nevertheless, comparing corresponding models, the improvement in fit from adding the exogenous shocks to weekly, post 1970 models is smaller than that from doing so to the post-1929 monthly models. This difference may reflect that bond markets were hit by larger shocks in the 1930s than in the last half century when the safety net and automatic stabilizers were also more extensive and important in cushioning the macroeconomy. To examine the impact of the Fed's corporate bond programs on the real macro-economy, we use our estimates to adjust bond spreads in forecasts using the recently revised Federal Reserve Board Model of the U.S. economy (FRBUS) described by LaForte (2018). In the FRBUS model, the equity risk premium is mainly driven by movements in the spread between yields on S&P BB-rated corporate and the 10-year Treasury bond. This spread is equivalent to our Moody's based Baa-Treasury spread. 13 The monthly model estimates plotted in Figure 7 imply that the announced interventions lowered the Baa-Treasury spread by 1.93 percentage points in 2020:q2, which are similar to what our weekly models imply (figure available upon request). In the FRBUS model, the spread between yields on investment-grade corporate and 10- year Treasury bonds affects the real economy via three main channels. First, a higher spread depresses stock prices by pushing up the equity risk premium. In addition, a higher spread raises the user cost of capital, thereby depressing business investment, and pushes up spreads of home mortgage over long-term Treasury interest rates, thereby lowering residential investment. To assess the impact of the corporate bond facilities on real GDP, we ran simulations of the FRBUS model under the assumptions of an inertial Taylor Rule and that the effective zero lower bound on the federal funds rate would be binding throughout the forecast period. Using publicly available inputs, these assumptions yield a baseline. Alternative assumptions for the BBB -10-year Treasury spread are generated by assuming the counterfactual that the Fed corporate bond programs were not implemented. Each of the two estimated models imply different monthly paths for the equilibrium spread, which for simplicity we assume would have immediately affected the actual spread absent the PMCCF and SMCCF. In each case, quarterly averages of the spread would have been higher in 2020:q2 and 2020:q3, and slightly higher in 2020:q4 before wearing off. This exercise produces somewhat lower paths for real GDP than the baseline that translate into more negative real output gaps (scaled by potential real GDP), which are reported in Table 4 . After four quarters (2021:q2), real GDP is 0.7 to 0.8 percentage points lower, with the largest effect in the monthly model. After 8 quarters, the magnitudes grow somewhat, ranging from 1.0 to 1.3 percentage points, and remain similar after 8 quarters. This reflects the fact that the forecast incorporates the limits that the zero lower bound puts on the policy response to a wider output gap. Pre-COVID recessions were driven by different factors than those driving the COVID recession. In turn, expectations of the persistence of the COVID recession could differ. Coupled with different policy responses, these can limit how much the analysis can attribute differences in the behavior of corporate spreads to the corporate bond programs. Nevertheless, the timing of movements in the unemployment rate and the Baa-Treasury spread were in line that of prior recessions until the corporate bond programs were announced. Even then, because several policy responses were implemented around the start of the COVID recession, our estimates may capture direct effects but not indirect ones and not fully account for general equilibrium effects. To partially address these concerns, this section provides more evidence on identifying corporate bond program effects and the robustness of our results to using other gauges of economic slack. Several programs or special policies occurred while the Fed's corporate bond programs were in effect. This raises the possibility that our aforementioned results may co-mingle the effects of the PMCCF and SMCCF with those of other policy actions, which fall into three categories. The first are programs providing liquidity, primarily in money markets. These include the Treasury provision of deposit insurance on money fund accounts, Fed backstop lending programs available to money funds and primary dealers, term discount loans, backstops or purchases of commercial paper (e.g., the Commercial Paper Funding Facility), and the Term Asset Backed Securities program that helped fund support for consumer and small business loans. However, these programs did not entail support for bonds issued by corporations. Furthermore, they were in place during the Great Recession and were revived at the start of the Covid Recession, and yet, the Baa-Treasury spread was very wide during the former episode but not during the latter. Another liquidity facility created during the Covid pandemic was the Municipal Liquidity firms that did not issue bonds, and the Fed's MainStreet Lending Program supported bank loans to firms ineligible for PPP loans and for the PMCCF and SMCCF. Hence, the firms directly affected were not issuing bonds whose pricing (spreads) could be plausibly and notably affected. The third set of interventions comprises the unconventional monetary policy tools of forward guidance (FG) and quantitative easing (QE). FG and often QE are explicitly intended to affect medium-and long-term Treasury (and sometimes mortgage) interest rates (Bernanke, 2021) . There are times of turmoil in financial markets, such as in March 2020 when Treasury yields rose because of a scramble for liquidity, and when QE was conducted by the FOMC -to widen spreads if they lower Treasury spreads relatively more than corporate yields ostensibly owing to the lower liquidity of the latter or reflecting that their use reflects expectations of a weaker economic outlook and higher defaults. To assess the impact of QE and FG on the Baa-Treasury spread we add controls for their short and long effects on Treasury bonds in a parsimonious way to limit the size of cointegrating vectors, which helps identify unique vectors. We added two long-term variables to the long-run models of the weekly spread and their first differences to the short-run equation of changes in the spread. One is the variable QEGDP, which is the ratio of seasonally adjusted weekly Federal Reserve assets to nominal potential GDP (quarterly CBO data are smooth and are interpolated into weekly readings). Adding the level of the ratio to the long-run equilibrium vector of the Baa spread controls for portfolio balance effects of QE, while adding the first differences to the equation of the change in the corporate spread controls for the pace of QE purchases. To parsimoniously control for forward guidance, the shift variable FG equals 1 in periods for which the Fed previously indicated that it would not change the federal funds rate (Yardeni, 2021) . These include the long stretch from December 16, 2008 until October 28, 2015 when it removed forward guidance, as well as the period since March 15, 2020. In addition, FG equals 1 during the period starting August 12, 2003, when the Fed announced that it anticipated maintaining its then low 1 percent federal funds rate target for a -considerable period,‖ and stating, until May 4, 2004, that it could be -patient‖ about when to begin raising the federal funds rate. Markets interpreted -considerable period‖ and -patient‖ as a commitment not to raise rates for a while (Carlstrom and Fuerst, 2005) . In practice, the Fed nearly simultaneously employed QE and FG in the last two decades. As a result, the measures of them, FG and QEGDP, are highly collinear making it difficult to obtain sensible long-run relationships and unique long-run relationships (cointegrating vectors) if both are added. To avoid these problems, we added each separately to weekly models from Table 3, with Tables 7 and 8 reporting results adding FG and QEGDP, respectively. In each of the latter two tables, models 1 and 2 repeat the pre-Covid Model 1 from Table 3 except that Model 2 includes one of measures of the two unconventional monetary tools. In Table 6 , Models 3 and 4 rerun Models 1 and 2 over a full sample ending in late March 2021 without adjustments for the Fed's corporate bond programs. In contrast, Models 5 and 6 rerun Models 1 and 2 over the full sample with adjustments to the squared unemployment rate for the Fed's corporate bond programs. As reported in Table 7 and 8, in each case-especially the models using a pre-Covid sample-the FG and QE variables are insignificant and do not affect the qualitative results regarding the need to adjust the models for the corporate bond programs. In models estimated through the full sample, the models that account for the corporate bond programs by capping the unemployment rate continue to have clean residuals and better model fits than the corresponding models that are not adjusted for the backstop effects of the Fed corporate bond programs. We also ran a placebo-test by replacing the Baa-Treasury spread with the 10-year Treasury yield. As shown in Appendix Table B , there was no substantial difference between corresponding placebo models that capped or did not cap the unemployment rate, whether in terms of finding a long-run relationship, estimates of long-run coefficients, and speeds of errorcorrection, nor in the fit or degree of serial correlation in the residuals of the short-run models. This is consistent with the view that Treasury yields and the provision of liquidity in that market were not the main channel through which policy actions limited the rise of corporate spreads in the COVID recession. As a robustness check, we reviewed alternative measures of slack of which two are available at a monthly but not weekly frequency and are available over a long sample. Monthly models of the Baa-Treasury spread were run using the GDP output gap or manufacturing capacity utilization to replace the squared unemployment rate as a gauge of slack. One advantage of the GDP output gap is that it more comprehensively measures economic slack and the risk it poses to companies and their debt holders than the unemployment rate. In addition, provided that the estimates of potential GDP are accurate, GDP output gaps better implicitly control for capacity than actual unemployment rates. A major disadvantage is that the quarterly frequency of standard GDP output gaps severely limits degrees of freedom to assess the Fed's corporate bond program. To address this, we model the monthly Baa-Treasury spread using monthly nominal GDP output gaps to replace monthly unemployment rates. IHS Markit Economics constructs a nominal monthly GDP series starting in 1992 that applies techniques to monthly input data mimicking those of the Bureau of Economic Analysis. To these, we spliced monthly nominal GDP estimates from Stock and Watson that span 1959:01-2010:06. 15 CBO quarterly estimates of potential nominal GDP are interpolated into monthly readings and the monthly GDP gap (GDPGap) is the difference between actual and potential GDP as a percent of potential. Using this 1959:01-2021:03 sample, we estimate the following specification of the level and the first difference of the Baa spread: which omit Accord and short-run Great-Depression shock variables that predate the sample. The long-run model in eq. (4a) slightly modifies eq. (2a) by using the level of the GDP gap 15 Over the 1992-2012 period of overlap, Stock and Watson series was 2.7 to 3.2 percent below the IHS Markit estimates and 2.7 to 3.0 below the latter in 1992. We used the 2.8 percent January 1992 difference to multiplicatively splice the series. These differences owe to benchmark revisions of real GDP and its price deflator. Splicing earlier Stock and Watson monthly real GDP ($2005) with current IHS Markit real estimates ($2012) entails more judgment as the combination of shifts in the base year and other revisions create less consistent differences in the overlap period. rather than the level multiplied by its absolute value and by not interacting the Penn-Central bankruptcy level shift dummy with the measure of slack. These changes yield better fitting models and reflect the different nature of the slack variables. For full sample, the GDP gap is used without adjustment in some models that ignore the Fed corporate bond program. For models accounting for these programs which capped the spread after late March 2020, we limit the absolute magnitude of the negative readings on the output gap over March-December 2020 to either the size of that month's negative gap (4.11 percent) or the actual gap. This adjusted GDP gap (GDPGapCCF) reflects that the programs effectively capped spreads from April to July 2020. Similarly, for capacity utilization, the program-adjusted series uses the March 2020 value of CapUtil as a minimum from April to December 2020, which was binding from April to August 2020. Table 5 use the monthly GDPGap specifications in equations. (4a) and (4b), and parallel those using the unemployment rate. Models 1, 3, and 4 omit the short-run dummy for Lehman's collapse, which is in Models 2, 5, and 6. Models 1 and 2 are pre-Covid models estimated through March 2020, while the others are estimated through 2021:03 and include two monthly dummies for the onset of Covid in April and May 2020. Models 3 and 5 use the unadjusted GDPGap series, while models 4 and 6 use the CCF-adjusted series. Corresponding models using monthly capacity utilization are presented in Table 6 , with the sample starting in January 1953. 16 Unique and significant cointegrating vectors are found for all models in Table 5 with significant coefficients on the long-term variables with a negative sign on the output gap variables and positive signs on PennCentral and CFMA. In Table 6 , using capacity utilization, unique and significant vectors are found for the two pre-Covid models and for full sample models adjusted for the corporate bond programs, while unique equilibrium relationships were not found in full sample models (3 and 5) that ignore the programs. Paralleling the output gap models, the long-term variables in Table 6 are significant with a negative sign on capacity utilization variables and positive signs on PennCentral and CFMA. In both tables, residuals are well-behaved in the two pre-Covid era models and the two full sample models that account for the Fed corporate bond programs, whereas residuals are plagued by serial correlation in the two full sample models (3 and 5) that ignore the programs. Consistent with Figures 5 and 6 , the equilibrium spreads implied by the pre-Covid models in Tables 5 and 6 overpredict spreads in months when the Fed's programs effectively backstopped spreads (April through mid-summer 2020). Overall, the qualitative findings are robust to replacing the unemployment rate with the alternative slack measures. The Covid-19 induced financial crisis and recession created unprecedented challenges for the Fed. The pandemic-induced declines in consumption, investment and labor hours were magnified by government mandated lockdowns in spring 2020. Key to the amplification of negative shocks in an earlier dramatic crisis and depression, 1929-33, was the financial accelerator which followed the financial collapse associated with serious banking panics. The consequent decline in net worth and collateral led to defaults, bankruptcies, and a collapse in credit. Early in the pandemic a similar dynamic occurred as firms were hit by a collapse in sales, orders and a disrupted supply chain. This stress is evident in a spike in the Baa-10-year Treasury bond spread. In response to the crisis, the Fed restarted facilities developed in the GFC of [2007] [2008] to restore the plumbing of the financial system and to bolster the banking system. In the recent crisis the Fed added new facilities to shore up the corporate, municipal government, and small to medium business sectors. It was able to do this because of explicit Treasury guarantees against credit losses, which were not made in the GFC. Key to the effort was the creation of the primary and secondary corporate credit facilities that were intended to support the issuance of, and trading in, corporate bonds, respectively, and at non-crisis spreads over Treasury yields. The announcement of these facilities halted a rapid rise in the Baa-Treasury bond spread that was intrain in March 2020 at the height of the crisis-and seven weeks before the first Fed purchase of corporate bond ETFs. As a result, rather than continuing to surge, this and other investmentgrade corporate spreads peaked at levels seen in more normal recessions, and the spreads have subsequently ebbed. This paper models the corporate Baa-Treasury spread using monthly data back to 1929 and weekly data since 1971, accounting for the major historical credit market shocks. Movements in the spread are highly correlated with the business cycles, as proxied by the square of the unemployment rate, but also by the GDP output gap and capacity utilization. These effects do not appear to stem from other policy actions, as indicated by our review of program timing or firm eligibility, and by evidence that the effects do not owe to QE or Fed forward guidance. Our models show that the announced Fed intervention largely capped the spike of the spread in March 2020. We also find based on the estimated effects on bond spreads, that the Later, in the GFC, it intervened in commercial paper. In both cases, its actions were confined to the money market, consistent with the traditional central bank objective of maintaining adequate liquidity to prevent and ameliorate financial panics. The recent corporate bond market interventions go beyond that traditional lender of last resort role. Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector. (ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 UR t-1 2 + α 2 PennCent t -1 x UR t-1 2 + α3PreAccord t-1 + α 4 CFMA t-1 ]. (iii) First difference terms of elements in the long-run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length. Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector. (ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 UR t-1 2 + α 2 CFMA t-1 ]. (iii) First difference terms of elements in the long-run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length and if a time trend is included in the long run. Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector. (ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 GDPGap t-1 + α 2 PennCent t -1 + α3CFMA t-1 ]. (iii) First difference terms of elements in the long-run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length. Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector.(ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 GDPGap t-1 + α 2 PennCent t -1 + α3CFMA t-1 ]. (iii) First difference terms of elements in the long run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length. Short-Run: ΔBaaTR t = β 0 + β 1 EC t-1 + ∑ n i=1 β 2i ΔBaaTR t-i + ∑ n i=1 β 3i ΔUR t-i 2 + ∑ n i=1 β 3i ΔCFMA t + ΩX t + μ t EC t-1, -0.018 ** -0.017 ** -0.018 ** -0.017 ** -0.017 ** -0.017 * Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector. (ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 UR t-1 2 + α 2 CFMA t-1 + α 3 FG t-1 ]. (iii) First difference terms of elements in the long run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length and if a time trend is included in the long run. Short-Run: ΔBaaTR t = β 0 + β 1 EC t-1 + ∑ n i=1 β 2i ΔBaaTR t-i + ∑ n i=1 β 3i ΔUR t-i 2 + ∑ n i=1 β 3i ΔCFMA t + ΩX t + μ t EC t-1, -0.018 ** -0.019 ** -0.018 ** -0.019 ** -0.017 ** -0.019 ** (4. Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a two-equation system with at most one cointegrating vector. (ii) EC t-1 = BaaTR t-1 -[α 0 + α 1 UR t-1 2 + α 2 CFMA t-1 + α 3 QEGDP t-1 ]. (iii) First difference terms of elements in the long-run cointegrating vector. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length and if a time trend is included in the long run. TR * t = α 0 + α 1 UR t 2 + α 2 PennCent t x UR t 2 + α3PreAccord t + α 4 CFMA t (monthly) TR * t = α 0 + α 1 UR t 2 + α 4 CFMA t (weekly) (Corresponding placebo models with and without CCF adjustments did not significantly differ in terms of longterm and error-correction coefficients, cointegration, and degree of serial correlation in model residuals.) Weekly ( (4) 3.88 7.41 99.01 ** 96.18 ** Notes: Absolute t-statistics are in parentheses and ** ( * ) denoted significant at the 99% (95%) confidence level. (i) Long-run: Maximum likelihood estimates of the long-run equilibrium relationship using a twoequation system with at most one cointegrating vector. (ii) Impact dummies and first difference terms of elements in the long-run cointegrating vector omitted to conserve space. (iii) Models 1 and 2 (3 and 4) correspond to models 5 and 6 in Table 2 (Table 4 ) with the modification that the lag length and number of weekly COVID impact dummies lengthened to 15 in models 1 and 2, which resolved serial correlation in the residuals for both models. (iv) Lag lengths chosen to minimize the AIC criterion. (v) significance of the trace and VEC Auto statistics reflects lag length. Duca 2. Data collection. John V. Duca 3. Data analysis and interpretation Liquidity Risk of Corporate Bond Returns: Conditional Approach The Asset-Backed Securities Markets, the Crisis, and TALF Lombard Street: A Description of the Money Market Non-Monetary Effects of Financial Crises in the Propagation of the Great Depression The Macroeconomics of the Great Depression: A Comparative Approach The New Tools of Monetary Policy The Real Effects of Disrupted Credit: Evidence from the Global Financial Crisis Financial Fragility and Economic Performance The Financial Accelerator and the Flight to Quality The Financial Accelerator in a Quantitative Business Cycle Framework.‖ In Handbook of Macroeconomics Board of Governors of the Federal Reserve System. 2020a. -Federal Reserve Issues FOMC Statement Board of Governors of the Federal Reserve System. 2020b. -Federal Reserve Issues FOMC Statement Board of Governors of the Federal Reserve System. 2021. -Timelines of Policy Actions and Communications: Forward Guidance about the Federal Funds Rate Should Derivatives Be Privileged in Bankruptcy? An Historical Perspective on the Crisis of Rules for a Lender of Last Resort: An Historical Perspective How New Fed Corporate Bond Programs Dampened the Financial Accelerator in the Covid-19 Recession.‖ NBER Working Paper 28097 How the New Fed Muni Bond Program Capped Muni-Treasury Yield Spreads in the Covid-19 Recession.‖ NBER Working Paper 28437 Economic Policy Uncertainty and the Credit Channel: Aggregate and Bank Level U.S. Evidence over Several Decades Fiscal and Financial Crises.‖ NBER Working Forthcoming. -The 1932 Federal Reserve Open Market Purchases as a Precedent for Quantitative Easing The Promise and Performance of the Federal Reserve as Lender of Last Resort 1914-1933.‖ NBER Working Paper It's What You Say and What You Buy: A Holistic Evaluation of the Corporate Credit Facilities The Impact of the Coronavirus (COVID-19) Pandemic on The Employment Situation for Establishing Credible Rules for Fed Emergency Lending Considerable Period of Time: The Case of Signaling Future Policy.‖ Federal Reserve Bank of Cleveland Economic Commentary New Highs in Unemployment Insurance Claims.‖ FRBSF Economic Letter Impacts of the Fed Corporate Credit Facilities Through the Lenses of ETFs and CDX.‖ Federal Reserve Bank of Chicago Working Paper What Credit Market Indicators Tell Us,‖ Federal Reserve Bank of Dallas Economic and Financial Policy Review Did the Commercial Paper Funding Facility Prevent a Great Depression Style Money Market Meltdown? The Money Market Meltdown of the Great Depression The Subprime Mortgage Crisis,‖ Federal Reserve System History Gateway Essay The Great Depression Versus the Great Recession in the U.S.: How Fiscal, Monetary, and Financial Policies Compare The Other (Commercial) Real Estate Boom and Bust: the Effects of Risk Premia and Regulatory Capital Arbitrage Housing Markets and the Financial Crisis of 2007-2009: Lessons for the Future Would a Bagehot Style Corporate Bond Backstop Have Helped Counter the Great Recession? American Default: The Untold Story of FDR, the Supreme Court, and the Battle over Gold Financial Fragility in the Covid-19 Crisis: the Case of Investment Funds in Corporate Bond Markets.‖ NBER Working Paper 27559 Recession Risk and the Excess Bond Premium The Debt Deflation Theory of Great Depressions A Monetary History of the United States 1867-1960 The Fed Takes on Corporate Credit Risk: An Analysis of the Efficacy of the SMCCF.‖ NBER Working Paper 27809 Credit Spreads and Business Cycle Fluctuations The Federal Reserve and Panic Prevention: The Roles of Financial Regulation and Lender of Last Resort The Federal Reserve and Panic Prevention: The Roles of Financial Regulation and Lender of Last Resort When Selling Becomes Viral: Disruptions in Debt Markets in the COVID-19 Crisis and the Fed's Response Generalizations of the KPSS-test for Stationarity.‖ Statistica Neerlandica Lender of Last Resort: The Concept in History.‖ Federal Reserve Bank of Richmond Economic Review Cyclical Variations in the Risk Structure of Interest Rates Likelihood-Based Inference in Cointegrated Vector Autoregressive Models The First Year: The Role of a Modern Lender of Last Resort Lender of Last Resort: A Contemporary Perspective Overview of the Changes to the FRB/US Model Alex) Zhou. 2020. -Mutual Fund Fragility, Dealer Liquidity Provisions, and the Pricing of Municipal Bonds.‖ Asymmetric Information and Financial Crises: A Historical Perspective.‖ In Financial Markets and Financial Crises Stock Market Crashes and Their Aftermath: Implications for Monetary Policy.‖ NBER Working Moody's Investor Service Anatomy of a Liquidity Crisis: Corporate Bonds in the COVID-19 Crisis The Political Economy of Prosperity 2020. -Current Economic Issues.‖ Speech at the Peterson Institute for International Economics The Limits of Arbitrage Derivatives and the Legal Origin of the 2008 Credit Crisis Why Was There No Banking Panic in 1920-1921? The Federal Reserve Banks and the Recession.‖ Unpublished manuscript The Treasury Market in Spring 2020 and the Response of the Federal Reserve.‖ BIS Working Paper 966 Chronology of Fed's Quantitative Easing.‖ Yardeni Research Stationarity ( Notes: * and ** denote 95% and 99% significance levels, respectively. Lag lengths for the KPSS stationarity tests are based on the Newey-West bandwidth selector using a Quadratic Spectral kernel for the spectral estimation method (see Hobjin, et al. 2004) . The combination of a significant KPSS stationary test statistic on the level of a variable (rejecting that it is stationary) and a significant test statistic on its first difference (accepting it is stationary) is evidence against trend stationarity. Unit root tests include a time trend for the unemployment rate variable combinations reflecting that there could be moving trends in the natural rate of unemployment. Tests omit a time trend for the output gap models which account for trends in potential output. The sample used for CapUtil is 1951:01-2021:03 reflecting the longer sample period and lagged first difference terms used in Table 6 .