The Financial Market Effects of the Federal Reserve's Large-Scale Asset Purchases

Questions & answers

Q1. What is the paper’s core question and headline finding?

The paper asks how the Federal Reserve’s first large-scale asset purchase program (LSAP1 / QE1, December 2008-March 2010) affected financial markets, and finds that it produced “economically meaningful and long-lasting reductions in longer-term interest rates on a range of securities, including securities that were not included in the purchase programs,” with these reductions coming “primarily” from lower risk premiums — especially term premiums — rather than from lower expected future short-term rates (Abstract, p. 3). This is the founding empirical study of what later became known as quantitative easing in the US context, combining two independent methods (an event study and a time-series regression) to triangulate the size of the effect.

Q2. Through what mechanisms do the authors argue the purchases affected interest rates?

The primary channel is portfolio balance: when the Fed buys a long-duration asset and pays for it by creating zero-duration bank reserves, it reduces the amount of duration risk the private sector must hold, and for investors to be willing to hold less of it, the purchased security’s price must rise (yield fall). Because the purchases remove duration broadly, not just from the specific securities bought, the resulting compression in term premiums spills over to imperfect substitutes the Fed never purchased (drawing on Tobin’s portfolio-balance framework and preferred-habitat models such as Vayanos-Vila 2009). The authors stress that “the portfolio balance effect has nothing to do with the expected path of short-term interest rates” — the Fed deliberately preserved its ability to raise short rates, so any long-rate decline had to come through risk premiums, not signaling (pp. 6-8). MBS purchases additionally reduced prepayment/negative-convexity risk. A second, distinct channel operated early in the program: purchases began when market liquidity was poor and spreads unusually wide, and the Fed’s presence itself improved market functioning and narrowed agency-debt and MBS spreads (pp. 8-10).

Q3. How large were the purchases, and what does the time-series data set look like?

Between December 2008 and March 2010 the Fed purchased more than $1.7 trillion in longer-term Treasury, agency debt, and agency MBS — about 22% of the $7.7 trillion stock of those securities outstanding at the start; measured in duration-adjusted “ten-year equivalents” the purchases were about $850 billion, more than 20% of the $3.7 trillion ten-year-equivalent stock (p. 11). The program was announced in two steps — November 2008 (up to $100bn agency debt plus $500bn agency MBS) and March 2009 (expanded to $1.75 trillion total, adding up to $300bn in longer-term Treasuries) — with purchases concentrated in the 2- to 10-year maturity sectors (pp. 4, 11-13). The time-series analysis uses monthly data from January 1985 to June 2008, with the 10-year Kim-Wright (2005) term premium as the dependent variable (a 10-year Treasury yield specification is used as a robustness check), and explanatory variables covering the business cycle (unemployment gap, core CPI), uncertainty (inflation disagreement, realized volatility), and the net public-sector supply of longer-term debt, both unadjusted and duration-adjusted (Section 4.3, pp. 26-28).

Q4. How does the event study identify the effect, and what does it find?

The event study measures cumulative one-day changes in financial variables around an eight-announcement “baseline event set” running from the initial LSAP announcement on November 25, 2008 through FOMC statements in August, September, and November 2009, and finds the 10-year Treasury yield fell 91 bp, the 10-year agency debt yield fell 156 bp, agency MBS yields fell 113 bp, the 10-year Kim-Wright term premium fell 71 bp, the 10-year swap rate fell 101 bp, and the Baa corporate index fell 67 bp, versus only a 34 bp decline in the 2-year Treasury yield (Table 1, Figure 5, pp. 19-20). The much larger decline at the 10-year than the 2-year horizon, together with the large drop in the Kim-Wright term premium itself, is read as evidence the announcements worked mainly by lowering the term premium rather than by signaling lower future policy rates; the substantial swap-rate and Baa-index declines show the effect reached well beyond the securities actually purchased (p. 20). The two FOMC statements in the event set that also contained forward-guidance language (December 16, 2008 and March 18, 2009) did not appear to drive the result — the Kim-Wright one-year-ahead instantaneous rate moved only about 4 bp on each of those dates (pp. 23-24). Overall the authors describe longer-term rates as having declined “by up to 150 bp” around the key announcements (p. 24).

Q5. What does the independent time-series regression find, and how does it identify the effect?

Using OLS and dynamic OLS (DOLS, following Stock-Watson 1993) to exploit a cointegrating relationship between the 10-year term premium and its determinants, the paper estimates that a 1-percentage-point-of-GDP increase in the net public supply of longer-term debt raises the 10-year term premium by about 4.4 bp (unadjusted) or 6.4 bp (duration-adjusted) in OLS, and by roughly similar magnitudes (~4 bp / ~6 bp) in DOLS, with cointegration confirmed (adjustment speed -0.15, half-life about five months) (Tables 2-4, pp. 29-31). Scaling this relationship up to the full $1.725 trillion LSAP program (about 12% of 2009 GDP) implies a total reduction in the 10-year term premium of about 52 bp (OLS, unadjusted) to 38 bp (OLS, duration-adjusted), with the DOLS estimates and a yield-level specification giving a wider overall range of roughly 38-82 bp across specifications (Table 5, p. 33). The key identifying assumption is that the net debt-supply variable is exogenous to the term premium; the authors note that if this is violated it would bias the estimated effect downward, making their estimates “somewhat conservative” rather than overstated (Section 4.3, p. 27).

Q6. How do the two independent approaches compare, and why does that matter?

The time-series estimate of the total LSAP effect on the 10-year term premium (about 38-82 bp) “overlaps considerably” with the event-study estimate, which the authors call impressive given that the two approaches use entirely separate data and methodologies (p. 33). This convergence is central to the paper’s evidentiary strategy: rather than relying on a single identification approach, whose validity depends on a specific set of assumptions, the authors treat agreement between an event study (which assumes a well-specified announcement set and an efficient, uncontaminated event window) and a time-series regression (which assumes exogenous debt supply) as corroborating evidence that the estimated term-premium effect is not an artifact of either method’s particular assumptions.

Q7. What does the international evidence from Japan and the UK show?

The UK’s 2009-10 gilt purchase program (about £200 billion, roughly 12% of UK GDP) is associated with a roughly 100 bp decline in 5- to 25-year gilt yields, which the authors describe as “strikingly similar” to the roughly 106 bp two-day decline in the US 10-year Treasury yield, alongside a 70 bp decline in UK investment-grade and 150 bp decline in UK speculative-grade corporate yields (Section 5, pp. 35-38, citing Joyce et al. 2010); one notable difference is that UK 10-year swap rates fell only about 10 bp versus about 100 bp in the US. By contrast, Japan’s 2001-06 quantitative easing program had only small effects on longer-term JGB yields, which the authors attribute to Japanese government bond holdings under the program peaking at only about 4% of GDP (versus roughly 12% in the US case), purchases skewed toward shorter residual maturities, and the Ministry of Finance’s decision to lengthen new JGB issuance, which offset the intended duration-removal effect (pp. 35-36, citing Ugai 2007 and McCauley-Ueda 2009).

Q8. What limitations and caveats do the authors themselves flag?

The authors describe the event study’s identifying assumptions as “strong”: that the chosen eight-event set captures all the news that moved LSAP expectations, that nothing else moved those expectations on the same days, that the event windows are wide enough to capture the full effect but narrow enough to exclude other news, and that markets are informationally efficient (fn. 20, p. 17) — and they note that isolating LSAP-specific news from other policy or economic-outlook news in the same announcements is inherently difficult. They also flag that the net 10-year Treasury yield did not actually decline over the full program window — it rose more than 100 bp on non-event days — but attribute this to a large rise in the expected fiscal deficit, an improving economic outlook, and the unwinding of the fall-2008 flight-to-quality, rather than to a reversal of the LSAP effect itself (pp. 20-22, Figure 6). The Kim-Wright term-premium regression is estimated on a pre-crisis sample that does not include a major financial crisis or a zero-lower-bound episode, which is part of why the authors add the yield-level specification as a robustness check (p. 31). Finally, the authors note that the program’s discrete, front-loaded design (decided in two steps with little subsequent responsiveness to conditions) may not be optimal, and they flag the open question of whether a more systematic policy rule for asset purchases would be preferable (Conclusion, pp. 39-40).