The Information Content of Capacity Utilisation Rates for Output Gap Estimates Michael Graff and Jan-Egbert Sturm 17 October 2010 Overview Introduction and motivation Data Output gap data: OECD Economic Outlook Capacity utilisation: information from Business Tendency Surveys Empirical analysis Design Results Conclusions 3rd International Seminar, Moscow 17 October 2010 2 Real-time data Vintage Final data Evaluation Final data Final Partly revised Partly revised First-released First-released Time Economic forecast Political decisions 3rd International Seminar, Moscow 17 October 2010 3 Measurement of the output gap in real time Output gap = (Y – Y*)/Y* ≈ y – y* • Percentage deviation of factual output from potential output Potential and factual output are unobservable in real time This is when this information is most needed as a guidance for economic and monetary policy • Countercyclical fiscal policy • E.g. Swiss “debt brake” • Monetary policy in a Taylor rule framework 3rd International Seminar, Moscow 17 October 2010 4 Measurement of the output gap in real time Problems with real-time estimates of output gap data Late availability and revisions to Y • All GDP data are either ex-post estimates, or real-time “nowcasts” or ex-ante forecasts End-point problem when estimating Y* Orphanides & Van Norden (2002) Revisions are of similar magnitude as the gap itself Hence, questionable usefulness of output gap data in real time • How can we improve the quality of output gap estimates in real time? Various remedies suggested • Forecasting data points • Multivariate filters • … • This paper: output gap capacity utilisation from BTS 3rd International Seminar, Moscow 17 October 2010 5 Some methods to estimate potential output Smoothing real GDP using a filter Hodrick-Prescott, Baxter-King, … The “split time trend” method calculate average output growth during each cycle, where the cycle is defined as the period between peaks in economic growth Estimating potential output using a production function approach ln Y = a + α ln L + (1 – α) ln K + TFP where L is labour, K capital and TFP ‘total factor productivity’ ln Y* = a + α ln L* + (1 – α) ln K* + TFP*, where ‘*’ denotes ‘potential’ 3rd International Seminar, Moscow 17 October 2010 6 Output gap data: OECD Economic Outlook Production function based approach Bi-annual vintages with data at annual frequency First vintage: Jun. 1995 (data cover 1970-1996) Last vintage: Dec. 2009 (data cover 1970-2011) Bi-annual vintages with data at quarterly frequency First vintage: Dec. 2003 (data cover 1970q1-2005q4) Last vintage: Dec. 2009 (data cover 1970q1-2011q4) The resulting revision data sets are unbalanced Annual data: 22 countries (up to 287 obs.) Quarterly data: 18 countries (up to 338 obs.) The largest balanced panels thereof are Annual data: 17 countries, 1996-2005 (170 obs.) Quarterly data: 14 countries, 2003q3-2005q4 (140 obs.) 3rd International Seminar, Moscow 17 October 2010 7 Capacity utilisation data Sources: European Commission, OECD MEI, KOF, national sources (in case of Belgium and New Zealand) Business tendency survey data Question asks for assessment of current level of capacity utilisation – Refers mainly to means of production (physical capital) – Is consistently asked in the industry sector – Range • Minimum: completely idle = 0 % • Maximum: full utilisation of present capacity = 100 % - Few surveys allow for “excess” capacity utilisation > 100% Data are (almost) not revised 3rd International Seminar, Moscow 17 October 2010 8 BTS: Direct measurement of capacity utilisation 3rd International Seminar, Moscow 17 October 2010 9 Australia Austria Belgium Czech Republic Denmark Finland France Germany Hungary Ireland Italy Japan Luxemburg Netherlands New Zealand Norway Poland Portugal Spain Sweden Switzerland United Kingdom No. countries Output gap (vintages) Capacity utilisation (reference period) 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 2005:Dec–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 2005:Dec–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 2005:Dec–2009:Dec 1995:Jun–2009:Dec 1997:Jun–2009:Dec 1995:Jun–2009:Dec 2006:Dec–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1995:Jun–2009:Dec 1996q1-2009q4 1996q1-2009q4 1980q1-2009q4 1993q2-2009q4 1987q1-2009q4 1993q1-2009q4 1985q1-2009q4 1985q1-2009q4 1996q1-2009q4 1985q1-2008q2 1970q1-2009q4 1978q1-2009q4 1985q1-2009q4 1985q1-2009q4 1970q1-2009q4 1987q1-2009q4 1992q2-2009q4 1987q1-2009q4 1987q2-2009q4 1996q1-2009q4 1970q1-2009q4 1985q1-2009q4 22 22 Annual data Countries in bold are not included in the strictly balanced sample 3rd International Seminar, Moscow 17 October 2010 10 Data setup and revision process: Annual data Reference Period Vintages / Release Dates 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec 1970 … 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 … R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 … … R7 R5 R3 R1 F3 F1 R8 R6 R4 R2 F4 F2 Fx Forecast number x Rx Release number x Source: OECD, calculations KOF 3rd International Seminar, Moscow 17 October 2010 11 Releases of annual output gaps: unbalanced panel 3rd International Seminar, Moscow 17 October 2010 12 Releases of annual output gaps: averaged bal. panel 2.0 % of potential GDP CU rate (in %) 83.5 1.5 83.0 1.0 82.5 0.5 82.0 0.0 81.5 -0.5 81.0 -1.0 80.5 -1.5 80.0 -2.0 79.5 1996 rel. 1 Source: OECD, calculations KOF 1997 rel. 2 1998 rel. 3 1999 2000 rel. 4 2001 rel. 5 2002 rel. 6 2003 rel. 7 2004 rel. 8 3rd International Seminar, Moscow 2005 CU rate 17 October 2010 13 Average absolute revisions of annual OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 14 Average revisions of annual OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 15 Cumulative revisions of annual OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 16 Descriptive Statistics of the annual releases/vintages degree (in %) Release 1 Release 2 Release 3 Release 4 Release 5 Release 6 Release 7 Release 8 Maximum panel Strictly balanced panel (22 countries, 1995-2009) (17 countries, 1996-2005) Obs Mean St.D. Min. Max. Obs Mean St.D. Min. Max. Capacity utilisation (in % of full capacity) 353 81.5 4.40 64.5 92.3 170 81.4 2.97 74.4 87.5 Output gap (in % of potential GDP) 287 -0.93 1.98 -8.79 5.50 170 -0.79 1.58 -4.86 5.50 283 -0.55 1.65 -5.73 5.68 170 -0.64 1.58 -4.27 5.68 287 -0.40 1.75 -5.50 6.39 170 -0.53 1.64 -4.31 6.39 283 -0.46 1.90 -7.31 6.41 170 -0.46 1.62 -4.06 6.41 287 -0.38 1.97 -7.32 7.66 170 -0.38 1.65 -4.53 7.66 283 -0.51 1.90 -9.54 6.77 170 -0.28 1.59 -3.16 6.77 287 -0.48 1.96 -9.54 6.84 170 -0.24 1.64 -5.11 6.84 283 -0.52 2.02 -9.66 6.83 170 -0.13 1.67 -4.17 6.83 3rd International Seminar, Moscow 17 October 2010 17 Estimation design Data revisions contain news revisions are orthogonal to earlier releases and not predictable yRx(t) = yR1(t) + (t), cov(yR1(t),(t)) = 0 • Rx = R2, R3, R4, R5, R6, R7, R8 Mincer-Zarnowitz (1969) test for forecast efficiency (in a panel data set-up) Are real time output gap estimates “informationally efficient” (w.r.t. Capacity Utilisation data) Are the revisions predictable? Rx-R1y(t) = (i) + yR1(i,t) + CU(i,t) + (t) + (i,t) – Rx-R1y(t) represent the cumulative revisions 1 to 7 – Hypotheses: (i) = 0 , = 0 , = 0 3rd International Seminar, Moscow 17 October 2010 18 Efficiency regressions, increasing revision horizons Dependent variable: First release (y R 1) Capacity utilisation rate 2 (1) (2) R2-R1 R3-R1 (3) R4-R1 (4) R5-R1 (5) R6-R1 (6) R7-R1 (7) R8-R1 -0.22 -0.35 -0.48 -0.55 -0.59 -0.54 -0.47 (-3.74) (-6.13) (-8.00) (-6.69) (-7.35) (-6.25) (-5.63) 0.06 0.13 0.17 0.19 0.23 0.16 0.16 (1.53) (2.50) (2.44) (2.66) (3.29) (1.99) (2.19) Adjusted R Number of observations Number of countries Number of periods 0.19 170 17 10 0.26 170 17 10 0.32 170 17 10 0.37 170 17 10 0.47 170 17 10 0.45 170 17 10 0.49 170 17 10 p-value LR-test for country effects p-value LR-test for time effects p-value LR-test for time and country effects 0.05 0.00 0.00 0.09 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3rd International Seminar, Moscow 17 October 2010 19 Goodness-of-fit across different revisions 0.50 adj.R2 0.45 0.40 0.35 0.30 0.25 0.20 0.15 Revision 1 Cumulative Revision 2 Cumulative Revision 3 Cumulative Revision 4 without CU variable Cumulative Revision 5 Cumulative Revision 6 Cumulative Revision 7 CU variable included 3rd International Seminar, Moscow 17 October 2010 20 Additional regression results, annual data Dependent variable: cumulative revision 7 (Δ R8-R1y) First release (y R 1) Capacity utilisation rate Capacity utilisation rate, lagged one period 2 (1) (2) (3) (4) (5) -0.39 -0.47 -0.49 -0.42 -0.48 (-6.09) (-5.63) (-6.00) (-5.98) (-5.91) 0.16 0.18 0.12 (2.19) (2.57) (1.90) 0.17 0.12 (2.52) (1.92) Adjusted R Number of observations Number of countries Number of periods 0.47 170 17 10 0.49 170 17 10 0.49 167 17 10 0.50 167 17 10 0.50 167 17 10 p-value LR-test for country effects p-value LR-test for time effects p-value LR-test for time and country effects 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 3rd International Seminar, Moscow 17 October 2010 21 Data setup and revision process: Quarterly data Reference Period Vintages / Release Dates 2003 2004 2005 2006 2007 2008 2009 Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec Jun Dec 1970 … 2003 2003 2003 2003 2004 2004 2004 2004 2005 2005 2005 2005 … 2010 I … I II III IV I II III IV I II III IV … IV … R2 R2 R1 R1 F4 F4 F3 F3 F2 F2 F1 F1 … R3 R3 R2 R2 R1 R1 F4 F4 F3 F3 F2 F2 … … R4 R4 R3 R3 R2 R2 R1 R1 F4 F4 F3 F3 … … R5 R5 R4 R4 R3 R3 R2 R2 R1 R1 F4 F4 … Fx Forecast number x Rx Release number x Source: OECD, calculations KOF … R6 R6 R5 R5 R4 R4 R3 R3 R2 R2 R1 R1 … … R7 R7 R6 R6 R5 R5 R4 R4 R3 R3 R2 R2 … … R8 R8 R7 R7 R6 R6 R5 R5 R4 R4 R3 R3 … … … … R8 R8 R7 R7 R6 R6 R5 R5 R4 R4 … R8 R8 R7 R7 R6 R6 R5 R5 … R8 R8 R7 R7 R6 R6 … … … R8 R8 R7 R8 R7 R8 … … F1 F2 Based on less information Based on more information 3rd International Seminar, Moscow 17 October 2010 22 Releases of quarterly output gaps: unbalanced panel 3rd International Seminar, Moscow 17 October 2010 23 Average absolute revisions of quarterly OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 24 Average revisions of quarterly OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 25 Cumulative revisions of quarterly OG, balanced panel 3rd International Seminar, Moscow 17 October 2010 26 Efficiency regressions, quarterly data 3rd International Seminar, Moscow 17 October 2010 27 Conclusions Revisions to OECD output gap estimates are almost of a similar magnitude as the output gap estimates During the (short) sample period, output gaps were overall revised upwards Hence, revisions appear to be predictable without further information Yet, in addition to this, real time BTS data on capacity add further explanatory power to explain revision process OECD real-time output gap estimates are not informationally efficient Referring to the survey data available in real time could have improved output gap estimates Findings are robust with respect to sample and frequency 3rd International Seminar, Moscow 17 October 2010 28 To be done … Carry out real-time forecasting exercise on country level Do the results hold when using other output gap estimates (e.g. those produced by HP filters) Do other BTS data, e.g. business climate indicators, add information … (suggestions are highly appreciated) 3rd International Seminar, Moscow 17 October 2010 29