Monetary Tightening Cycles and the Predictability of Economic Activity. by Tobias Adrian and Arturo Estrella * October 2006.

Transcription

1 Monetary Tghtenng Cycles and the Predctablty of Economc Actvty by Tobas Adran and Arturo Estrella * October 2006 Abstract Ten out of thrteen monetary tghtenng cycles snce 1955 were followed by ncreases n unemployment, three were not. The term spread at the end of these tghtenng cycles dscrmnates between the two subsequent outcomes, but the levels of nomnal or real nterest rates do not. JEL Classfcaton: E44, E52, G0 Keywords: Monetary polcy, nterest rates, term structure, dscrmnant analyss, logt * Both authors: Captal Markets Research Functon, Federal Reserve Bank of New York, 33 Lberty Street, New York, NY Correspondng Author: Tobas Adran, tobas.adran@ny.frb.org, tel , fax Presented at Bundesbank/ZEW conference, November 23-24, We are grateful to Stavros Perstan for useful suggestons and to Thomas Laubach for provdng detaled results from Laubach and Wllams (2003). The vews expressed n ths paper are those of the authors and do not necessarly represent those of the Federal Reserve Bank of New York or the Federal Reserve System.

2 1 Classfyng monetary tghtenng cycles The conventonal objectve of monetary polcy tghtenng s to ncrease the cost of borrowng, whch tends to slow down real actvty and hence nflatonary pressure. The extent to whch future real actvty s reduced depends on the degree of tghtenng and s dffcult to gauge n real tme. In ths paper, we nvestgate the ablty of fnancal ndcators to dscrmnate between tghtenng cycles that are followed by declnng real actvty and those that are not. In partcular, we nvestgate the forecastng power of the term spread, levels of nomnal and real federal funds rates, and the dfference between the real federal funds rate and ts long-run equlbrum value. We nvestgate monetary tghtenng cycles snce 1955, for whch we can obtan consstent monthly data. In general terms, we defne the end of a tghtenng cycle as the month n whch the federal funds rate peaks, after steady ncreases over a perod of sx months to one year. More precsely, we assume cycles end when ether of these two crtera s met: (1) the federal funds rate s hgher than at any tme from 12 months before to 9 months after and s at least 50 bass ponts hgher than at the begnnng of ths perod, or (2) the federal funds rate s hgher than at any tme from 6 months before to 6 months after and s 150 bass ponts hgher than the average at these endponts. The applcaton of these rules leads to reasonable results, as shown n Fgure 1. The ends of cycles are ndcated by vertcal lnes and NBER recessons by shadng. Over the last 50 years, we dentfy thrteen tghtenng cycles. The frst crteron (longer wndow) by tself dentfes most of the cycles, but msses three (Aug. 1971, Sept. 1973, Apr. 1980) that nvolve substantal ncreases n the funds rate. Two of these three were followed by recessons. 1

3 Our datng of the ends of monetary cycles generally follows the chronology of the begnnngs of tghtenng cycles from Romer and Romer (1989), although we tend to dentfy more cycles. 1 Each Romer date wthn our sample perod s followed drectly by a cycle end date, wth the lone excepton correspondng to two consecutve Romer dates (Aug and Oct. 1979) between whch the federal funds rate essentally dd not fall. 2 Forecastng real actvty at the end of tghtenng cycles Is t possble to antcpate the evoluton of real actvty followng the endngs of tghtenng cycles? We nvestgate the ablty of fve fnancal ndcators to forecast the response of the real economy to monetary tghtenng. The level of the nomnal federal funds rate as a measure of monetary polcy stance s proposed by Bernanke and Blnder (1992) and Bernanke and Mhov (1998), who examne ts usefulness n an dentfed VAR framework (together wth controls for nflaton). Laubach and Wllams (2003) propose the gap between the current real nterest rate and the natural rate of nterest as measure of monetary tghtness. We use three alternatve measures of the real federal funds rate: adjusted by CPI nflaton over the last 12 months, adjusted by expected core PCE nflaton, as n Laubach and Wllams (2003), and the gap between the latter and the Laubach-Wllams equlbrum real rate. Fnally, we use the spread between the 10-year constant maturty Treasury rate and the bond-equvalent secondary market 1 Romer and Romer (1989) classfy the begnnngs of tghtenng cycles between 1947 and 1988 by nvestgatng the mnutes of the Federal Open Market Commttee for language that sgnals a reversal n monetary polcy. They dentfy Oct. 1947, Sep. 1955, Dec. 1968, Apr. 1974, Aug. 1978, Oct. 1979, and Dec as begnnngs of monetary cycles. 2

4 rate on 3-month Treasures, whch Estrella and Hardouvels (1991) and others have shown forecasts recessons well. All nterest rates are monthly averages of daly data. We use two measures of subsequent real actvty: conventonal NBER turnng pont dates and the maxmum cumulatve ncrease n the unemployment rate n the 18 months followng the end of each tghtenng cycle. NBER recesson dates are wdely used and do not requre much dscusson. The unemployment rate measure avods the mplct dscreton n the NBER datng, relyng nstead on a mechancal rule. Each measure of actvty s converted nto a dummy by askng whether a recesson ensued wthn 18 months of the end of the tghtenng cycle or whether the unemployment rate ncreased over the same perod. In Table 1, we lst the dates of the tghtenng cycles together wth our fnancal and real ndcators. 2 Among the thrteen monetary tghtenng cycles over the past 50 years, only three dd not lead to an ncrease n unemployment n the 18 months followng the peak: Aug. 1971, Aug. 1984, and Apr The other ten tghtenng cycles were followed by an ncrease n unemployment and n all but one case by an NBER-dated recesson. We see n the last two columns that the only dscrepancy between the unemployment and NBER ndcators s after Nov. 1966, a perod that has been called a credt crunch or a mn-recesson by many and an actual recesson by Fredman (1968). In Fgure 2, we plot the (maxmal) ncrease n the unemployment rate after each monetary peak aganst the term spread durng the month of the peak for each of the twelve cases snce In the fgure, an ntrgung pattern emerges. The three peaks that were not followed by an ncrease n unemployment were accompaned by a term spread of 125 bass ponts or 2 Note that data avalablty precludes the calculaton of the frst two observatons for the two ndcators based on core PCE nflaton. 3

5 more. The remanng ten fed funds peaks were accompaned by a term spread below 35 bass ponts at the tme of the polcy reversal. In Fgure 3, we plot the relatonshps between other nterest rates measures at the end of tghtenng cycles and subsequent changes n unemployment. Ths fgure clearly suggests that none of the other fnancal ndcators s helpful n classfyng the response of real actvty to monetary polcy tghtenng. To confrm ths, however, we apply two formal statstcal technques. 3 Statstcal analyss Dscrmnant analyss seems lke a natural method to apply to our problem. We would lke to use a fnancal ndcator x (where runs over the n = 13 ends of tghtenng cycles) to classfy the cases nto one of two populatons, one n whch real actvty slows down and one n whch t does not. Let y {0,1} be an ndcator of an economc slowdown, based on ether NBER dates or the rse n unemployment. Dscrmnant analyss provdes a rule of the form: classfy an observaton as y = 1 f f( x ) > 0 and otherwse as y = 0. When x s a vector of ndcators, the sample dscrmnant functon s 1 1 ( ˆ π ˆ π ) ˆ ˆ ( ˆ μ ˆ μ ˆ μ ˆ μ ) ( ˆ μ ˆ μ ) f ( x) = log 1 2 Σ Σ + where ˆ π j s the sample frequency of y and ˆ 1 Σ x, (1) = j, ˆ μ j s the sample mean of x condtonal on y ˆ 1 Σ= ˆ ˆ ˆ ˆ + n ( x μ1)( x μ1) ( x μ0)( x μ0). 2 y = 1 y = 0 = j, 4

6 When x s scalar, the functon (1) s monotonc and the dscrmnant condton may be expressed as an nequalty n x. Our second measure s based on a logstc regresson of the form ( ) ( ˆ ˆ 1 β0 β1 ) P y = = F + x, (2) where F s the cumulatve logstc dstrbuton. Efron (1975) shows that dscrmnant analyss s typcally, though not always, more effcent than logstc regresson n classfyng samples from two populatons. We therefore present n Table 2 results based on both measures. Consder frst the case of the term spread, wth real actvty defned n terms of NBER recessons. The dscrmnant condton for classfyng an end of tghtenng as a slowdown s x < A look at the values of the varables n Table 1 shows that the only observaton not classfed correctly s Nov It s known from the lterature that ths s the one false postve frequently encountered when usng the term spread to predct NBER recessons. The spread also does well when gauged by the logt standard, especally n relatve terms, wth an R- squared of 55%. The other four fnancal ndcators fare much worse by ether statstcal technque. In each case, the dscrmnant condton cannot sort out the dfferences, classfyng all the observatons as slowdowns. The logt results lead to a smlar concluson, although the nomnal and the PCEadjusted federal funds rates have R-squared slghtly above 10%. The real funds rate gap does a bt worse, whch may be a consequence of the hgh level of uncertanty n the estmaton of the equlbrum rate, as reported by Laubach and Wllams (2003). Results usng the unemployment rate as a measure of real actvty are qualtatvely smlar. The one salent dfference s that the term spread has a perfect record usng ether 5

7 dscrmnant or logt analyss. In contrast, the logt R-squared for the other ndcators s somewhat worse than when NBER recessons are used to represent real actvty. The statstcal reason for the relatve success of the term spread s smple. Its range of values when unemployment subsequently rses s to 0.31%, as compared wth 1.25 to 1.82% when unemployment declnes. Not only s there no overlap, but there s a substantal gap between the two ranges. In contrast, the non-recessonary observatons for each of the other varables are nterspersed among the recessonary cases, as Fgure 3 shows. For nstance, recessonary fgures for the real funds rate (CPI) have been as low as 0.63% and as hgh as 9.84%, whch encompasses the range of 1.29 to 7.43% for the non-recessonary cases. Note fnally that classfcaton rules other than f( x ) > 0 are possble, such as rules that cap the probablty of one type of classfcaton error. For nstance, when the unemployment ndcator s used, the rule that classfes y = 1 when the term spread s less than 2 bass ponts lmts the probablty of msclassfyng an expanson as a recesson to 5%. Smlarly, the rule that y = 1 when the term spread s less than 90 bass ponts lmts the probablty of msclassfyng a recesson as an expanson to 5%. More generally, there s no guarantee that the future performance of the term spread wll match the hstorcal record snce It seems clear from the evdence, however, that ts potental usefulness as a leadng ndcator n perods of monetary tghtenng should not be overlooked. 6

8 References Bernanke, B.S., and A.S. Blnder, 1992, The federal funds rate and the channels of monetary transmsson, The Amercan Economc Revew 82(4), Bernanke, B.S., and I. Mhov, 1998, Measurng monetary polcy, Quarterly Journal of Economcs 113(3), Efron, B., 1975, The effcency of logstc regresson compared to normal dscrmnant analyss, Journal of the Amercan Statstcal Assocaton 70(352), Estrella, A., 1998, A new measure of ft for equatons wth dchotomous dependent varables, Journal of Busness and Economc Statstcs 16(2), Estrella, A. and G. Hardouvels, 1991, The term structure as a predctor of real economc actvty, Journal of Fnance 46, Fredman, M., 1970 Controls on nterest rates pad by banks, Journal of Money, Credt and Bankng 2(1), Laubach, T. and J.C. Wllams, 2003, Measurng the natural rate of nterest, Revew of Economcs and Statstcs (85), Romer, C.D. and D.H. Romer, 1989, Does monetary polcy matter? A new test n the sprt of Fredman and Schwartz, NBER Macroeconomcs Annual 4,

9 Table 1: End of Monetary Tghtenng Dates and Fnancal and Real Indcators Real Fed Funds Rate (Expected Real Fed Funds Gap (Expected End of Tghtenng Date Fed Funds Rate Real Fed Funds Rate (Lagged CPI) PCE) PCE) 10-Year / 3- Month Spread Subsequent Change n Unemployment Oct NA NA Nov NA NA Nov Aug Aug Sep Jul Apr Jun Aug Mar Apr Jul NBER Recesson Indcator Notes: All varables are expressed n percent, except for the dchotomous NBER ndcator. The real fed funds gap s computed by subtractng from the real PCE-adjusted rate the Laubach-Wllams (2003) one-sded estmate of the equlbrum real rate for the quarter n whch the monthly observaton falls. 8

10 Table 2: Statstcal Analyss of the Relaton between Interest Rates and Real Actvty Measure of Real Actvty: NBER Recessons Increase n Unemployment Rate Indcator Dscrmnant condton Correctly classfed Logt R- Squared Dscrmnant condton Correctly classfed Logt R- Squared Term Spread < / < / Fed Funds Rate > / > / Real Fed Funds Rate (CPI) > / < / Real Fed Funds Rate (Core > / > / PCE) Real Fed Funds Gap (PCE) > / > / Note: Dscrmnant condtons are expressed n percent. Logt R-squared s the Estrella (1998) measure of ft. 9

11 Fgure 1: The Fed Funds Rate, Ends of Tghtenng Cycles (grd), and NBER Recessons (shadng) Percent

12 Fgure 2: The 10-year mnus 3-month spread and subsequent unemployment ncreases : : : :10 u ncrease, % : : :07 Dscrmnant condton: spr < : : : : : : year 3-month spread, % 11

13 Fgure 3: The Fed Funds Rate and Subsequent Unemployment Changes u ncrease, % u ncrease, % u ncrease, % Dscrmnant condton: ff > : : : : : : : : Fed funds rate, % Real fed funds rate (CPI), % 1966: : : : : :08 Dscrmnant condton: rff gap > : : : : : : : : : : : : : : : : : :04 Dscrmnant condton: rff < Real fed funds rate gap (PCE), % 1984: : : : :06 12