Monetary Policy Alternatives at the Zero Bound: An Empirical Assessment

The paper asks whether “nonstandard” monetary policies remain effective once the short-term policy rate is at or near the zero lower bound, and answers empirically — using “the methods of modern empirical finance” — rather than relying on theory alone, drawing on the recent experience of the United States and Japan. It organizes nonstandard policy into three classes, following Bernanke and Reinhart (2004): (1) using communications to shape expectations about the future path of the policy rate; (2) increasing the size of the central bank’s balance sheet (quantitative easing); and (3) changing the composition of the balance sheet (targeted asset purchases). Two complementary empirical strategies are used throughout: an event-study/factor-decomposition approach for communications, and a no-arbitrage affine term-structure model for quantity and composition effects.

A Cholesky decomposition of three market-based indicators around 116 FOMC decisions (July 1991-2004, one-hour windows from 15 minutes before to 45 minutes after the announcement) separates a “current policy surprise” factor from a second “path” factor tied to the statement’s language, and the path factor dominates: it explains roughly four-fifths of the variance of the year-ahead Eurodollar futures rate (versus about one-fifth for the current-surprise factor) and 68% of the variance of the five-year Treasury yield (with the current-surprise factor at 12% and a third factor at 20%) (Table 1, p. 30). This confirms the two-factor finding of Gürkaynak, Sack, and Swanson. Statements that were surprising raised the variance of the path factor sharply — about 196 basis points versus about 42 basis points when the statement was as expected (Table 2, p. 34) — and signed surprises moved year-ahead policy expectations by about 12 basis points on average (16 basis points for path surprises specifically), with a hawkish statement raising the five-year yield by roughly 8 basis points (10 for a path surprise), holding the current-policy surprise fixed (Table 3, pp. 36-37).

After the Fed introduced “considerable period” language in August 2003, the ten-year Treasury yield’s sensitivity to payroll surprises roughly tripled, from about 4 basis points per 100,000 jobs (August 1991-August 2003) to about 11 basis points per 100,000 jobs (September 2003-August 2004) (Figure 2, p. 39). Using the term-structure model, the authors compute an implied upper-bound effect of the “considerable period” language on the yield curve itself, ranging from about 25 basis points at the two-year horizon to about 7 basis points at the ten-year horizon, based on VAR shocks that lowered the Eurodollar futures rate by 63 basis points (Figure 5, p. 49).

A five-factor no-arbitrage affine term-structure model (a VAR with 4 lags, monthly, June 1982-2004, driven by the employment gap, core PCE inflation, expected inflation, the federal funds rate, and the year-ahead Eurodollar futures rate) generates a counterfactual benchmark for the yield curve given macro conditions and the stance of policy; residual deviations of actual from predicted yields are then attributed to factors outside the model, such as relative-supply effects. Applied to the 1999-2002 Treasury debt buyback, the twenty-year yield fell from about 20 basis points above to about 80 basis points below the model’s benchmark, with a bootstrap ex ante probability under 10% (statistically significant); the effect was concentrated at the long end with no material effect at short maturities, and 30-year swap spreads widened while 2-year spreads did not — suggesting limited pass-through of the supply effect to private-sector rates (Figures 6-8 and Table 6, pp. 52-55).

Around Japanese Ministry of Finance foreign-exchange interventions (January 2000-March 2004), US two-, five-, and ten-year Treasury yields fell significantly — by about 0.78 to 0.83 basis points per $1 billion intervened on all days, and a smaller 0.55 to 0.66 basis points excluding major-data-release days — while T-bill yields were unaffected; five- and ten-year yields ran 50-100 basis points below model predictions over the period (Table 7, p. 57; Figure 9). Separately, during the May-June 2003 “deflation scare,” when targeted Treasury purchases were anticipated, term-structure model errors spiked down by about 50 basis points or more, reversing in July as the ten-year yield moved sharply around specific FOMC statements and Greenspan testimony (Table 8, p. 61); an illustrative put-option calculation suggests a commitment to buy the ten-year zero-coupon bond at a 5.5% yield one year hence would lower today’s ten-year rate by about 34 basis points, more than the 17 basis points by which the option is in the money, reflecting added convexity value (p. 62).

No — the paper finds little evidence the BOJ used statements to move near-term (second-factor) policy expectations, calling this “the strongest and most important contrast” with the Federal Reserve (Tables 10-11, pp. 67-70). There is, however, some evidence that quantity and composition actions moved asset prices directly: surprise increases in the JGB-purchase target are associated with the five-year JGB yield falling by about 5 basis points (a small effect from few observations), and surprise increases in the current-account-balance target are associated with the Nikkei 500 rising 3-6% on three of six occasions, including a 5.9% rise on the day quantitative easing was announced (pp. 69-71).

Even after imposing the zero lower bound by simulation (a shock pulls the policy rate back to zero whenever it would otherwise go negative), actual Japanese government bond yields ran below the levels the four-factor term-structure model predicted, and this gap widened after the zero-interest-rate policy began in February 1999 and again after quantitative easing began in March 2001 — giving the authors “some reason to believe that nonstandard policies in Japan have been effective at lowering longer-term interest rates” (Figures 10-11, pp. 73-76). The authors note this exercise “still has some shortcomings,” since the VAR dynamics and the pricing of risk may themselves change near the bound (fn. 91, p. 74).

The authors call their evidence “to some degree indirect,” because the US policy rate “has remained at least 100 basis points above zero” throughout the sample, so the paper never observes a true US ZLB episode (pp. 25, 77). They describe the size and reliability of quantity/composition effects as “quantitatively quite uncertain,” stating that “considerable uncertainty remains about the size and reliability of these effects under the circumstances prevailing near the ZLB” and that they “cannot be sure that the effects shown here are scalable in a predictable manner” (pp. 5, 55). Specific identification problems include the 2003 deflation scare overlapping with the period of MOF purchases and independently conveying deflation-risk information (p. 60), and potential joint endogeneity in the intervention regressions (p. 57). The Japanese analysis additionally rests on a small sample and a coarser two-day event window (versus one hour for the US) because intraday Japanese data are scarce (pp. 63-69). Pass-through to private borrowing rates may also be limited, since 30-year swap spreads widened during the Treasury buyback and a put-option-style rate peg would not pass through to corporate yields unless Treasuries and corporates are close substitutes (pp. 55, 62).

The authors conclude that the evidence — while “to some degree indirect” — generally supports the view that nonstandard policies would be effective at the zero lower bound: the Fed has used communications to affect expectations and longer-term yields, there is some evidence that relative asset supplies matter for US yields (a necessary condition for targeted asset purchases to work), and the Japanese term-structure results suggest longer-term yields were lower than otherwise expected under the zero-interest-rate policy and quantitative easing (Conclusion, pp. 77-78). Despite this, the authors say they “remain cautious about relying on such approaches” given the uncertainty documented above, and recommend that the best approach for a central bank is avoidance — maintaining a buffer against disinflation and easing preemptively so as to minimize the risk of reaching the zero lower bound in the first place (pp. 5, 77-78).