GDPNow – An Early Indicator of Economic Activity ‐Pat Higgins AUBER 2015 Fall Conference October 12, 2015 Disclaimer: The views expressed here are the presenter’s and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System. • GDPNow is a statistical model that nowcasts real GDP, and its subcomponents for the next quarter for which the U.S. Bureau of Economic Analysis has not yet released a published GDP estimate. • It is uses 100s of time series and it’s nowcasts can change after most major economic data releases. • The current and historical model nowcasts are all available on the Atlanta Fed’s website, along with a spreadsheet with all of the numerical details – source data and model parameters – showing exactly how the most recent nowcast is built up. Sources for GDP nowcasts Source Timing/frequency Atlanta Fed GDPNow 13‐14 updates a month; 5‐6 releases a month CNBC Rapid Update About 12 releases per month Survey of Professional Quarterly (70‐75 days before Forecasters (SPF) GDP release) GDP subcomponents # of panelists Individual forecasts available Free? 13 components (complete partition) None 7 components (complete partition) #N/A #N/A Yes About 10 No Yes 40 to 45 Yes Yes About 70 Yes Yes Wall Street Journal Forecasting Survey Middle of each month (72, 42 and 12 days before GDP) Blue Chip Economic Indicators 10th of each month (77, 47 and Consumption, net 17 days before GDP) exports, inventories. About 50 No, but average of top and bottom 10 forecasts. No Blue Chip Financial Forecasts 1st of each month (86, 56 and 26 days before GDP) About 45 Yes if 56 or 86 days before GDP release. No None None Other resources not listed: Bloomberg Monthly Forecasting Survey, Consensus Forecasts (Monthly), proprietary forecasts Evolution of Atlanta Fed GDPNow real GDP forecast for 2015: Q3 Quarterly percent change (SAAR) 4.5 Aug 14th: Industrial production (Spike in motor vehicle assemblies boosts inventories) 4.0 Sept 28th: Personal Consumption Expenditures Range of top 10 and bottom 10 forecasts 3.5 3.0 Blue Chip consensus 2.5 CNBC Rapid 2.0 Update survey 1.5 1.0 Atlanta Fed GDPNow forecast 0.5 0.0 28-Jul Sept 29th: Advance foreign trade report 4-Aug 11-Aug 18-Aug 25-Aug 1-Sep 8-Sep 15-Sep 22-Sep 29-Sep Date of forecast Sources: Blue Chip Economic Indicators, Blue Chip Financial Forecasts and CNBC/Moody’s Rapid Update Survey 6-Oct Atlanta Fed GDPNow forecasts for 2015: Q3, growth rates and changes Date 6-Aug 28-Aug 11-Sep 15-Sep 16-Sep 17-Sep 21-Sep 24-Sep 28-Sep 29-Sep 1-Oct 2-Oct 5-Oct 6-Oct 9-Oct Major Releases Initial nowcast Personal income and outlays Monthly Treasury Statement Retail sales/Industrial production Retail inventories (Sep 15), CPI Housing starts Existing-home sales Dur. maunf., New-home sales/costs Personal inc./outlays, GDP (Sep 25) Advance foreign trade (goods) ISM Manuf., Construction spending Employment, M3 Manufacturing Autos sales, ISM Nonmanufacturing International trade Wholesale trade, Imp/Exp prices GDP 0.9 1.2 1.3 1.5 1.4 1.5 1.4 1.4 1.8 1.1 0.9 0.7 0.9 1.1 1.0 EquipPCE ment 2.9 7.5 2.6 8.6 2.6 7.6 3.2 7.2 3.2 7.2 3.2 7.2 3.2 7.2 7.6 3.2 9.0 3.5 3.5 9.1 3.4 8.7 3.4 8.2 3.6 9.2 3.6 11.2 3.6 11.2 Intell. Change prop. Nonres. Resid. in net Change prod. struct. inves. Govt. Exports Imports exp. in CIPI 5.6 -0.6 5.4 1.0 3.3 4.7 -13 -68 6.8 2.4 9.4 1.5 4.3 5.8 -15 -63 6.8 4.2 7.4 0.8 3.3 3.7 -6 -61 6.8 4.8 6.9 0.8 3.0 3.3 -6 -69 6.8 4.8 6.8 0.8 3.0 3.3 -6 -71 6.8 4.7 9.9 0.8 3.0 3.3 -6 -71 6.8 4.7 7.6 0.8 3.0 3.3 -6 -71 6.8 4.9 8.4 0.8 3.0 3.3 -6 -74 7.0 5.1 8.5 0.8 3.0 3.5 -70 -7 7.0 5.1 8.5 0.8 1.3 6.8 -70 -37 7.0 2.8 6.1 0.9 1.1 6.5 -37 -70 6.9 2.8 6.1 1.0 1.0 6.4 -36 -75 6.9 2.8 6.2 1.0 1.0 6.5 -36 -76 6.9 2.8 6.2 1.0 1.9 6.1 -29 -76 6.9 2.8 6.2 1.0 2.0 6.2 -30 -80 Note: CIPI is “change in private inventories”. Changes in net exports and CIPI are both in billions of 2009 dollars (SAAR). All other numbers are quarterly percent changes (SAAR). Atlanta Fed GDPNow forecasts for 2015: Q3, contributions to growth Date 6-Aug 28-Aug 11-Sep 15-Sep 16-Sep 17-Sep 21-Sep 24-Sep 28-Sep 29-Sep 1-Oct 2-Oct 5-Oct 6-Oct 9-Oct Major Releases Initial nowcast Personal income and outlays Monthly Treasury Statement Retail sales/Industrial production Retail inventories (Sep 15), CPI Housing starts Existing-home sales Dur. maunf., New-home sales/costs Personal inc./outlays, GDP (Sep 25) Advance foreign trade (goods) ISM Manuf., Construction spending Employment, M3 Manufacturing Autos sales, ISM Nonmanufacturing International trade Wholesale trade, Imp/Exp prices GDP 0.9 1.2 1.3 1.5 1.4 1.5 1.4 1.4 1.8 1.1 0.9 0.7 0.9 1.1 1.0 PCE 2.00 1.77 1.75 2.16 2.15 2.16 2.15 2.15 2.38 2.37 2.32 2.30 2.40 2.41 2.41 Intell. Equip- prop. Nonres. Resid. Net ment prod. struct. inves. Govt. exports 0.22 -0.02 0.18 0.17 0.43 -0.31 0.50 0.27 0.07 0.30 0.26 -0.35 0.44 0.27 0.12 0.24 0.15 -0.14 0.42 0.27 0.13 0.23 0.15 -0.13 0.42 0.27 0.13 0.22 0.15 -0.13 0.42 0.27 0.13 0.32 0.15 -0.13 0.42 0.27 0.13 0.25 0.15 -0.13 0.44 0.27 0.13 0.27 0.15 -0.13 0.52 0.28 0.14 0.28 0.14 -0.15 0.52 0.28 0.14 0.28 0.14 -0.87 0.50 0.28 0.08 0.20 0.16 -0.86 0.47 0.28 0.08 0.20 0.18 -0.85 0.53 0.28 0.08 0.21 0.18 -0.86 0.64 0.28 0.08 0.21 0.18 -0.69 0.28 0.08 0.21 0.18 0.64 -0.70 CIPI -1.73 -1.61 -1.57 -1.76 -1.82 -1.82 -1.82 -1.90 -1.80 -1.79 -1.79 -1.91 -1.95 -1.95 -2.05 The current GDPNow forecast for third-quarter growth is about the same as the initial forecast made on August 6th. Upgrades to consumption and equipment investment have been offset by downgrades to net exports and inventory investment. Revisions to the third-quarter forecasts for GDP’s subcomponents have been relatively modest. Note: CIPI is “change in private inventories.” All numbers are percentage point contributions to GDP growth (SAAR). This is anHigh informal schematic howGDPNow the oil model and consumption BVAR following slides level flow chart ofofhow forecast of real GDPresults growthon is the assembled. lead me to think about the impact of oil prices on real spending of PCE subcomponents. Model forecast of real GDP growth in 2015q3. Currently 1.0 percent. Simple arithmetic 13 quarterly real GDP subcomponents for 2014q2‐ 2015q2 “Bayesian vector autoregression model” BVAR(5). A standard statistical forecasting model in economics. Forecast of high‐level GDP subcomponents (e.g. residential investment). Simple arithmetic Forecasts of very detailed GDP subcomponents (e.g. investment in manufactured homes). “Bridge equation” forecast April through September monthly source data for GDP that are already available Unreleased August and September monthly source data for GDP that need to be forecasted. Forecast from “factor augmented autoregression” Important monthly variables measuring economic activity: Nonfarm payroll employment, industrial production, ISM manufacturing index, many others. Note: The text in red are inputs to the model. The text in blue are outputs of the model. Factor‐model estimates of economic momentum Indices (Average=0, Standard deviation=1) 4 3 Chicago Fed National Activity Index (First principal component) 2 1 0 -1 -2 -3 -4 -5 GDPNow smoothed estimate of dynamic factor -6 -7 1967 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007 2011 2015 Dynamic factor model (Giannone, Reichlin and Small 2008): Estimated with Kalman filter/smoother Standardized growth rates of 126 real macro time series. Large intersection with data used to estimate Chicago Fed National Activity Index. 0.42 [Observer equation] 0.27 0.15 [State equation] Factor augmented autoregressions (Stock and Watson 2002): Forecast September’s real value of private nonresidential construction put in place (VPIP) 1200Δlog 0.43 6.05 1200Δlog Autoregressive coefficients sum to 0.31. Bridge equation (Klein and Sojo 1989, Chin and Miller [Minneapolis Fed] 1996): Predict 3rd quarter growth rate of real nonresidential structures investment excluding mining and wells 400Δlog : 6.1 0.36 0.96 400log 6.0 The chart below illustrates that the growth rate for real business structures investment excluding mining and wells is very highly correlated with the growth rate constructed from monthly source data. Since we get most of the latter before the GDP release we can make a fairly accurate forecast of the former using a simple “bridge equation”. Real investment and construction spending on business structures excluding mining and wells Quarterly, Annualized logarithmic percent change 40 Value of private nonresidential construction put in put in place (VPIP) deflated by weighted average of Producer Price Indexes for construction 30 20 10 0 -10 -20 Forecast* -30 Real investment in business structures excluding mining and wells -40 -50 Correlation between two series over 1985q1‐2015q2 period: 0.97 -60 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Weighted sum of forecasts Weights, estimated by restricted least squares, must sum to 1.0. Note: Annualized logarithmic growth rate is 400 log. Weights, estimated by restricted least squares, sum to 1.0. The weights on the monthly “bridge equation” models generally go up as we approach the date of the GDP release. Note: All forecasts are annualized quarterly log growth rates [400 log] except inventories. We have maintained real‐time forecasts of GDPNow since 2011q3. GDPNow forecasts have been more volatile than the Blue Chip consensus over this period. Blue Chip consensus forecasts have been slightly more accurate on average than GDPNow. Improvements have been made to GDPNow since 2011 (particularly with inventory and net exports forecasts) and after 2011q4, GDPNow has been slightly more accurate than the Blue Chip survey. Forecast errors of initial BEA estimates of quarterly GDP growth Actual and final forecasted values of initial BEA estimates of quarterly GDP growth Quarterly percent change (SAAR) 6 Percentage points 3 GDPNow model (last forecast) 5 4 Blue Chip Economic Indicators (last forecast) “Advance” estimate of GDP growth 2 Average absolute forecast error since 2011:Q3 Blue Chip: 0.61 ppt. GDPNow : 0.66 ppt. Root mean square forecast error since 2011:Q3 Blue Chip: 0.77 ppt. GDPNow : 0.89 ppt. Blue Chip Economic Indicators (last forecast) 1 3 2 0 1 -1 0 GDPNow model (last forecast) Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 2011 2012 2013 2014 2015 -2 -1 2011201220132014201520162017201820192020202120222023202420252026 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 Q3 Q4 Q1 Q2 2011 201220132014201520162017201820192020202120222023202420252026 2011 2012 2013 2014 2015 The BEA decomposes annualized quarterly real GDP growth into contributions from its subcomponents % ⋯ A similar decomposition is done with GDPNow and, with some modest assumptions, can be applied to either Blue Chip Economic Indicators or Survey of Professional Forecasters GDP growth nowcasts as well % Therefore, one can relate the average squared GDP growth forecast error over the 2011q3‐2015q2 period to the forecast errors of component contributions with: 1 16 % % Mean square forecast error of Real GDP growth 1 16 Mean square forecast error of contribution i 2 1 16 Mean cross product of errors of ith and jth contributions 13 Fixed effects regression to estimate average accuracy of GDP nowcasts for individual Wall Street Journal Economic Forecasting Survey panelists. Squared forecast error of ith panelist % , % , Time fixed effect: Normalized to sum to 0 over 2011q3‐2015q2 sample period Individual fixed effect for ith panelist Fixed effects regression to estimate average disagreement with WSJ survey consensus GDP nowcast for individual Wall Street Journal Economic Forecasting Survey panelists. % , % , Squared forecast difference between WSJ survey consensus and ith panelist , GDP nowcasts: Accuracy and distance from consensus forecast of both GDPNow and individual panelist forecasts from the Wall Street Journal Forecasting Survey over the 2011q3‐2015q2 period. Average accuracy Individual panelists Average accuracy Accuracy predicted from discrepancy WSJ survey consensus forecast WSJ survey consensus forecast Individual panelists Accuracy predicted from discrepancy Atlanta GDPNow Atlanta GDPNow Average discrepancy Average discrepancy Average accuracy • Individual panelists Accuracy predicted from discrepancy WSJ survey consensus forecast Since 2011q3, individual panelists from the WSJ survey whose forecasts are closest to the consensus (on average) tend to be among the most accurate (on average). • Average discrepancy Atlanta GDPNow GDPNow forecasts bucks this trend • Compared to other WSJ forecasters, it has tended to be relatively far away from the consensus forecast. • But it has been about as accurate as the average individual forecast. It has been less accurate the consensus forecast. GDPNow – An Early Indicator of Economic Activity ‐Pat Higgins AUBER 2015 Fall Conference October 12, 2015 Disclaimer: The views expressed here are the presenter’s and not necessarily those of the Federal Reserve Bank of Atlanta or the Federal Reserve System.