Nowcasting. Domenico Giannone Université Libre de Bruxelles and CEPR

Nowcasting Domenico Giannone Université Libre de Bruxelles and CEPR 3rd VALE-EPGE Global Economic Conference Business Cycles Rio de Janeiro, May 2013

Nowcasting Contraction of the terms Now and Forecasting Meteorology Nowcasting forecasting up to 6-12 hours ahead (long tradition, since 1860)

Nowcasting Contraction of the terms Now and Forecasting Meteorology Nowcasting forecasting up to 6-12 hours ahead (long tradition, since 1860) Only recently introduced in economics: Evans, 2005 IJCB; Giannone, Reichlin and Small, JME 2008 Why? Key variables released at low frequency and with long publication delays. Example, US GDP: Advanced estimate 4 weeks after end of the reference quarter Example, EA GDP: Flash estimate 6 weeks after end of the reference quarter

This Presetation 1 Nowcasting and the Real-Time Data-Flow with M. Bańbura, M. Modugno and L. Reichlin Prepared for the Handbook of Economic Forecasting, Volume 2, Elsevier-North Holland 2 Nowcasting China with S. Miranda Agrippino and M. Modugno In progress 3 Nowcasting Brazil M. Modugno In progress

Macroeconomic Forecasting and Conjuctural Analysis predicting the future: formal economic modeling of the relations among key macroeconomic aggregates predicting the present: simplified heuristic scrutiny of a variety of conjunctural data forecasts are updated infrequently disregarding the high-frequency flow of conjectural information quarterly updates (SPF and Central Banks), bi-annual updates (OECD, IMF) Research questions How important is nowcasting relative to longer horizon forecasting?

How important is nowcasting relative to longer horizon forecasting? Very!!!! Forecasting GDP in real time MSFE relative to constant growth Horizon 0 1 2 3 4 GB 0.87 1.03 1.16 1.23 1.29 SPF 0.85 1.03 1.00 1.06 1.06 Evaluation sample 1992Q1 through 2001Q4 The present is the only horizon of predictability. Unpredictability beyond current quarter Accuracy of macroeconomic projections heavily rely on starting conditions

How important is expert Judgement?

The Experts!

The Computer Nerde

Now-Casting US GDP: 10 years of Experience

Learning from the Markets Market participants can be viewed as now-casters they obsessively monitor all macroeconomic data to get a view on current and future fundamentals and their effects on policy

Learning from the Markets Market participants can be viewed as now-casters they obsessively monitor all macroeconomic data to get a view on current and future fundamentals and their effects on policy The relevant information on the state of the economy is conveyed to markets through the release of macroeconomic reports. Market expectation for the headlines of these reports are collected up to the day before the actual release of the indicator and distributed by data providers (i.e. Bloomberg). When realizations are different than these expectations, that is when the news are sizeable, the view of the market changes and this leads to changes in asset prices see (Boyd, Hu, and Jagannathan, 2005; Flannery and Protopapadakis, 2002)).

Learning from the markets: Bloomberg.com

Learning from the Markets Market participants can be viewed as now-casters they obsessively monitor all macroeconomic data to get a view on current and future fundamentals and their effects on policy The relevant information on the state of the economy is conveyed to markets through the release of macroeconomic reports. Market form expectations for the headlines of these reports up to the day before the actual release of the indicator and distributed by data providers (i.e. Bloomberg). When realizations are different than these expectations, that is when the news are sizeable, the view of the market changes and this leads to changes in asset prices see (Boyd, Hu, and Jagannathan, 2005; Flannery and Protopapadakis, 2002)).

The Real-Time Data-Flow: Last Week...

The Real-Time Data-Flow: News

Employment report on March 10, 2012 From the Bloomberg News: "Employers in the U.S. took on more workers than forecast in February, [...]" "[...] Stocks rose, capping the fourth straight weekly rally for the Standard & Poor s 500 Index [...]" "[...] Yields on benchmark 10-year note climbed to the highest in a week yesterday after the job report reduced speculation Federal Reserve policy makers may hint at next week s meeting that they re moving closer to more monetary stimulus[...]"

Employment report on August 3, 2012 From Bloomberg News: "The U.S. economy generated more jobs than forecast in July [...]" "[...] The S&P 500 advanced 1.9 percent to 1,390.99 in New York after dropping 1.5 percent in the previous four days [...]" "[...] The yield on the 10-year Treasury note climbed to 1.57 percent from 1.48 percent late yesterday[...]"

Employment report on September 7, 2012 Weak jobs report fuels QE3 hopes "[...] U.S. stocks wavered Friday, as investors digested a disappointing August jobs report but hoped the weakness would prompt the Federal Reserve to jumpstart the slowing economy with stimulus when it meets next week. [...]" "[...] The Labor Department reported that 96,000 jobs were added in August, well below the 120,000 jobs that the economists surveyed by CNNMoney were looking for.[...]" "[...] The Dow Jones industrial average and the Nasdaq edged up 0.1%, while the S&P 500 gained 0.3%. [...]"

Employment report on December 7, 2012 Forbes "Private payroll jobs increased by 146,000 in November above the 93,000 economists expected and the 12-month average of 157,000 per month.[... ] What does this report mean for traders and investors?" "[...] an improving economy will put the pressure on the Fed to launch another round of QE3. This is certainly bad news for precious metals. [...]" "[...]a stronger economy is bad news for Treasuries, especially at a time when they trade at record low yields. So, I will stay away from U.S. Treasuries [...]" "[...] investors may want to buy into economically sensitive sectors - like homebuilding - as higher employment increases the likelihood of people buying homes. [...]"

Mimicking Market behavior and Beyond (a) Construct a joint model for all macroeconomic data releases (b) Update the model in real time, in accordance with the real-time data flow Model based forecasts: free of judgement, mood, heading. Translate the news in a common unit What s the impact of the news on GDP?

A model of Now-Casting y Q t : GDP at time t. Ω v : vintage of data (quarterly, monthly, possibly daily) available at time v (date of a particular data release)

A model of Now-Casting y Q t : GDP at time t. Ω v : vintage of data (quarterly, monthly, possibly daily) available at time v (date of a particular data release) Nowcasting of yt Q : orthogonal projection of yt Q information: ] E [y t Q Ω v, on the available The information set Ω v has particular characteristics: 1 it has a ragged or jagged edge [publication lags differing across series] 2 it contains mixed frequency series, in our case monthly and quarterly 3 it could be large

Further features Projections need to be updated regularly ] ] E [y t Q Ω v, E [y t Q Ω v+1,... v, v + 1,..., consecutive data releases Typically the intervals between two consecutive data releases are short (possible couple of days or less) and change over time. Consequently, v has high frequency and it is irregularly spaced.

News and nowcast revisions New release the information set expands (new releases): Ω v Ω v+1 [we are abstracting from data revisions]

News and nowcast revisions New release the information set expands (new releases): Ω v Ω v+1 [we are abstracting from data revisions] Decompose new forecast in two orthogonal components: ] ] ] E [y t Q Ω v+1 = E [yt Q Ω v + E [yt Q I v+1, }{{}}{{}}{{} new forecast old forecast revision I v+1 information in Ω v+1 orthogonal to Ω v If we have a model that can account for joint dynamics of all variables, we can express the forecast revision as a weighted sum of news from the released variables: [ ] [ ] E yt Q Ω v+1 E yt Q Ω v }{{} forecast revision = j J v+1 b j,t,v+1 For detailed derivation see Banubra and Modugno, 2008. ( E [ xj,tj,v+1 x Ω j,tj,v+1 v]). }{{} news

What kind of framework? Three desiderata: 1 can capture joint dynamics of inputs and target 2 can be estimated on many series while retaining parsimony 3 can handle jagged edged data and mix frequency Idea: use parsimonious model that can be cast in state space form and use Kalman filter to project and to handle jagged edged data Evans 2005 IJCB; Giannone, Reichlin and Small, 2008 JME

Why Large? The Real-Time Data-Flow: This Week

Why Large? The Real-Time Data-Flow: Next Week

Computing projections. What kind of model? The dynamic factor model x t = µ + Λf t + ε t, f t : (unobserved) common factors; ε t : idiosyncratic components Λ factor loadings Factors are modelled as a VAR process: f t = A 1 f t 1 + + A p f t p + u t Parsimonious and robust model for Big Data - Forni et al. (2000), Stock and Watson (2002, Bernanke and Boivin (2002), Bai (2003), Giannone et al (2005) Alternative: Mixed Frequency Vector Autoregression + Use Shrinkage (BVAR) to make it work with Big Data. Schorfheide and Song (2011), McCracken, Owyang, Sekhposyan (2013), Ghysels (2012), Felsenstein, Funovits, Deistler, Anderson, Zamani, Chen (2013)

Problems and solutions Missing data Kalman filter and smoother can be used to obtain, in an efficient and automatic manner, the projection for any pattern of data availability in Ω v as well as the news I v+1 and expectations needed to compute b j,t,v+1 Mixed frequency Consider lower frequency variables as being periodically missing Estimation: Quasi Maximum likelihood: - robust and feasible Doz, Giannone and Reichlin., 2008 REStat - handling missing data Banbura and Modugno, 2010

State space representation with mixed frequencies Example: Let Yt Q denote the vector of (log of) the quarterly flow series. We assume that Y Q t is the sum of daily contributions X t Y Q t = t s=t k+1 X s, t = k, 2k,.... Hence we will have that the stationary series yt Q written as: y Q t = k t s=t k+1 t + 1 s k x s + t k s=t 2 (k 1) = Y Q t s t + 2 k 1 k x s Yt k Q can be where x s = X s X s 1 can be thought of as an unobserved daily growth rate (or difference). See also Modugno 2011: Nowcasting Inflation, t = k, 2k,.

The real-time data flow No Name Frequency Publication delay (in days after reference period) 1 Real Gross Domestic Product quarterly 28 2 Industrial Production Index monthly 14 3 Purchasing Manager Index, Manufacturing monthly 3 4 Real Disposable Personal Income monthly 29 5 Unemployment Rate monthly 7 6 Employment, Non-farm Payrolls monthly 7 7 Personal Consumption Expenditure monthly 29 8 Housing Starts monthly 19 9 New Residential Sales monthly 26 10 Manufacturers New Orders, Durable Goods monthly 27 11 Producer Price Index, Finished Goods monthly 13 12 Consumer Price Index, All Urban Consumers monthly 14 13 Imports monthly 43 14 Exports monthly 43 15 Philadelphia Fed Survey, General Business Conditions monthly -10 16 Retail and Food Services Sales monthly 14 17 Conference Board Consumer Confidence monthly -5 18 Bloomberg Consumer Comfort Index weekly 4 19 Initial Jobless Claims weekly 4 20 S&P 500 Index daily 1 21 Crude Oil, West Texas Intermediate (WTI) daily 1 22 10-Year Treasury Constant Maturity Rate daily 1 23 3-Month Treasury Bill, Secondary Market Rate daily 1 24 Trade Weighted Exchange Index, Major Currencies daily 1

Daily factor, GDP and its common component 3 Daily Factor GDP growth 2 1 0 03/01/1983 03/01/1984 03/01/1985 03/01/1986 03/01/1987 03/01/1988 03/01/1989 03/01/1990 03/01/1991 03/01/1992 03/01/1993 03/01/1994 03/01/1995 03/01/1996 03/01/1997 03/01/1998 03/01/1999 03/01/2000 03/01/2001 03/01/2002 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2009 03/01/2010-1 -2-3

Filter uncertainty, GDP 0.016 0.08 0.014 0.07 0.012 0.06 0.010 0.05 0.008 0.04 0.006 0.03 0.004 0.02 0.002 0.01 0.000 0.00 01/10/2008 08/10/2008 15/10/2008 22/10/2008 29/10/2008 05/11/2008 12/11/2008 19/11/2008 26/11/2008 03/12/2008 10/12/2008 17/12/2008 24/12/2008 31/12/2008 07/01/2009 14/01/2009 21/01/2009 28/01/2009 D W M Q Filter Uncertainty (rhs)

Forecasting the Great Recession 0.25 0 0.2-0.2 0.15-0.4 0.1-0.6 0.05-0.8 0-1 -0.05-1.2-0.1-1.4-0.15-1.6-0.2-1.8-0.25-2 01/10/2008 08/10/2008 15/10/2008 22/10/2008 29/10/2008 05/11/2008 12/11/2008 19/11/2008 26/11/2008 03/12/2008 10/12/2008 17/12/2008 24/12/2008 31/12/2008 07/01/2009 14/01/2009 21/01/2009 28/01/2009 D W M Q Fcst (rhs) Outturn (rhs)

Does information help improving forecasting accuracy? Root Mean Squared Forecast Error (RMSFE) 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 Q0 M1 D7 Q0 M1 D14 Q0 M1 D21 Q0 M1 D28 Q0 M2 D7 Q0 M2 D14 Q0 M2 D21 Q0 M2 D28 Q0 M3 D7 Q0 M3 D14 Q0 M3 D21 Q0 M3 D28 Q+1 M1 D7 Q+1 M1 D14 Q+1 M1 D21 Q+1 M1 D28 Benchmark Monthly BCDC Bridge STD SPF

S&P 500 and its common component at different levels of time aggregation 15 Daily growth rates 15 Month on month growth rates 10 10 5 5 0 5 10 0 5 10 15 15 20 20 25 25 1985 1990 1995 2000 2005 2010 30 1985 1990 1995 2000 2005 2010 15 Quarter on quarter growth rates 30 Year on year growth rates 10 20 5 0 10 5 0 10 10 15 20 20 25 30 30 40 35 1985 1990 1995 2000 2005 2010 50 1985 1990 1995 2000 2005 2010

Now-Casting and the Real-Time Data-Flow Research questions How important is nowcasting relative to longer horizon forecasting? Can we predict the present? How relevant is informal judgement? How relevant is the conjectural information? How often should we update the predictions? What have we learned Nowcasting is key! Little predictability beyond current quarter! We can predict the present without the need of informal judgement?. It is worth to obsessively monitor conjectural information! Accuracy improves significantly and continuously.

Now-Casting China In progress, with S. Miranda-Agrippino and M. Modugno

Now-Casting China The Data

Now-Casting China: The Common Factor

Now-Casting China in Real-Time

Now-Casting China: What are the Most Informative Data Releases

Now-Casting China In progress, M. Modugno

Now-Casting in Brazil An Award-Winning Central Bank

Now-Casting Brazil: The Data

Now-Casting China: What are the Most Informative Data Releases

Now-Casting Brazil: The Data