Macroeconomic Forecasting Using Diffusion Indexes
Summary (3 min read)
1. INTRODUCTION
- Recent advances in information technology make it possible to access in real time, at a reasonable cost, thousands of economic time series for major developed economies.
- The performance of these methods ultimately rests on the few variables that are chosen.
- This idea has a long tradition in macroeconomics.
- Forecasting is carried out in a two-step process: rst the factors are estimated (by principal components) using Xt , then these estimated factors are used to forecast ytC1.
- The improvement over the benchmark forecasts can be dramatic, in several cases produc- © 2002 American Statistical Association Journal of Business & Economic Statistics April 2002, Vol. 20, No. 2 147 ing simulated out-of-sample mean square forecast errors that are one-third less than those of the benchmark models.
2.1 An Approximate Dynamic Factor Model
- The authors begin with a discussion of the statistical model that motivates the DI forecasts.
- (The different time subscripts used for y and X emphasize the forecasting relationship.).
- The main advantage of this static representation of the dynamic factor model is that the factors can be estimated using principal components.
2.2 Estimation and Forecasting
- First, the sample data 8Xt9 T tD1 are used to estimate a time series of factors (the diffusion indexes), 8bFt9TtD1.
- Under a set of moment conditions for 4…1 e1F5 and an asymptotic rank condition on å, the feasible forecast is asymptotically rst-order ef cient in the sense that its mean square forecast error (MSE) approaches the MSE of the optimal infeasible forecast as N 1 T ! ˆ, where N D O4T 5 for any >.
- The assumptions by Stock and Watson (1998) are similar to assumptions made in the literature on approximate factor models (Chamberlain and Rothschild 1983; Connor and Korajczyk 1986, 1988, 1993), generalized to allow for serial correlation.
- A related dynamic generalization and estimation (but not forecasting) results were discussed by Forni, Hallin, Lippi, and Reichlin (2000).
2.3 Data Irregularities and Computational Issues
- In their dataset, some series contain missing observations or are available over a diminished time span.
- In these cases standard principal components analysis does not apply.
- The expectationmaximization (EM) algorithm can be used to estimate the factors by solving a suitable minimization problem iteratively.
- Ft could include lags of the dynamic factors ft , estimation of Ft might be enhanced by augmenting a vector of distinct time series with its lags.
- Xt with its lags, in which case the principal components of the stacked data vector are computed.
3.1 Forecasting Models and Data
- The forecasting experiment simulates real-time forecasting for eight major monthly macroeconomic variables for the United States.
- Four of these eight variables are the measures of real economic activity used to construct the Index of Coincident Economic Indicators maintained by the Conference Board (formerly by the U.S. Department of Commerce): total industrial production (ip); real personal income less transfers ; real manufacturing and trade sales ; and number of employees on nonagricultural payrolls .
- The price indexes are modeled as being I(2) in logarithms.
- I(1) speci cations also provide adequate descriptions of the data, particularly in the early part of the sample.
- In particular these variables performed well in at least one of the historical episodes considered by Staiger, Stock, and Watson (1997) (also see Stock and Watson 1999).
3.2 Simulated Real-Time Experimental Design
- Estimation and forecasting was conducted to simulate realtime forecasting.
- The rst simulated out of sample forecast was made in 1970:1.
- To construct this forecast, the data were screened for outliers and standardized, the parameters and factors were estimated, and the models were selected, using only data available from 1959:1 through 1970:1.
- (The rst date for the regressions was 1960:1, and earlier observations were used for initial conditions as needed.).
- All parameters, factors, and so forth were then reestimated, information criteria were recomputed, and models were selected using data from 1959:1 through 1970:2, and forecasts from these models were then computed for yh197022Ch .
4.1 Forecasting Results
- The results for the real variables are reported in detail in Table 1 for 12-month-ahead forecasts, and summaries for 6- and 24-month-ahead forecasts are reported in Table 2.
- The rst is the MSE of the candidate forecasting model, computed relative to the MSE of the univariate autoregressive forecast (so the autoregressive forecast has a relative MSE of 1.00).
- The simulated Table 1. Simulated Out-of-Sample Forecasting Results: Real Variables, 12-Month Horizon Industrial production Personal income Mfg & trade sales Nonag.
Stacked balance panel
- Finally, similar rankings of methods are obtained using I(1) forecasting models, rather than the I(2) models used here, that is, when rst rather than second differences of log prices are used for the forecasting equation and factor estimation.
- In their judgment, the performance of the leading indicator models reported here overstates their true potential out of sample performance, because the lists of leading indicators used to construct the forecasts were chosen by model selection methods based on their forecasting performance over the past two decades, as discussed in Section 3.
- The authors consider the performance of the various diffusion index models to be particularly encouraging.
Benchmark models
- This suggests that a very small state vector may be necessary for forecasting macroeconomic time series.
- These results raise several issues for future empirical and theoretical research.
- First, classical diffusion indexes are computed using nonlinear transformations of the data, but their indexes are linear functions of the data.
- The authors considered only U.S. data, and it would be useful to study the relative forecasting performance of these methods for other countries.
4.2 Empirical Factors
- Because the factors are identi ed only up to a k k matrix, detailed discussion of the individual factors is unwarranted.
- Nevertheless, the nding that good forecasts can be made with only one or two factors suggests brie y characterizing the rst few factors.
- These R2 are plotted as bar charts with one chart for each factor.
- Broadly speaking, the rst factor loads primarily on output and employment; the second factor on interest rate spreads, unemployment rates, and capacity utilization rates; the third, on interest rates; the fourth, on stock returns; the fth, on in ation; and the sixth, on housing starts.
5. DISCUSSION AND CONCLUSIONS
- The authors nd two features of the empirical results surprising and intriguing.
- First, only six factors account for much of the variance of their 215 time series.
- One interpretation of this Table 4. Simulated Out-of-Sample Forecasting Results: Price In‘ ation, 6- and 24-Month Horizons CPI Consumption de‘ ator CPI exc. food & energy Producer price index.
Did you find this useful? Give us your feedback
Citations
[...]
9,919 citations
3,170 citations
2,906 citations
2,866 citations
Cites methods or result from "Macroeconomic Forecasting Using Dif..."
...We have described the variables in detail in earlier work (Stock and Watson 2002)....
[...]
...AN EMPIRICAL EXAMPLE In related empirical work we have applied factor models and principal components to forecast several macroeconomic variables (see Stock and Watson 1999, 2002)....
[...]
...We have already reported similar results for other real macroeconomic variables (Stock and Watson 2002)....
[...]
...We gave details of the speci cation, including lag length choice and exact description of the variables, in earlier work (Stock and Watson 2002)....
[...]
1,540 citations
Cites background or methods from "Macroeconomic Forecasting Using Dif..."
...The second DGP is motivated by recent work on the predictive content of factor indexes of economic activity for output growth (examples include Stock and Watson, 2002, 2004; Marcellino et al., 2003; Shintani, 2005)....
[...]
...1 References include Lettau and Ludvigson (2001), Stock and Watson (2002, 2003, 2004) , Goyal and Welch (2003), Marcellino et al. (2003), Diebold and Li (2006), Orphanides and van Norden (2005), Rapach and Weber (2004), Clark and McCracken (2005b) and Shintani (2005)....
[...]
...The second DGP is motivated by recent work on the predictive content of factor indexes of economic activity for output growth (examples include Stock and Watson, 2002, 2004; Marcellino et al., 2003; Shintani, 2005)....
[...]
...But we refer to the quantiles of a certain non-standard distribution studied in Clark and McCracken 1References include Lettau and Ludvigson (2001), Stock and Watson (2002, 2003, 2004), Goyal and Welch (2003), Marcellino et al. (2003), Diebold and Li (2006), Orphanides and van Norden (2005), Rapach…...
[...]
References
2,866 citations
1,832 citations
1,166 citations
1,111 citations
Related Papers (5)
Frequently Asked Questions (7)
Q2. What is the main advantage of this static representation of the dynamic factor model?
The main advantage of this static representation of the dynamic factor model is that the factors can be estimated using principal components.
Q3. What is the definition of the estimated factors?
One interpretation of the estimated factors is in terms of diffusion indexes developed by NBER business cycle analysts to measure common movement in a set of macroeconomic variables, and accordingly the authors call the estimated factors diffusion indexes.
Q4. What is the form of the Phillips curve in a forecast?
The Phillips curve in ation forecasts considered here have the form (3.4), where Wt consists of the unemployment rate (LHUR) and mƒ1 of its lags, the relative price of food and energy (current and one lagged value only), and Gordon’s (1982) variable that controls for the imposition and removal of the Nixon wage and price controls.
Q5. What is the definition of a new frontier in macroeconomic forecasting?
This raises the prospect of a new frontier in macroeconomic forecasting, in which a very large number of time series are used to forecast a few key economic quantities, such as aggregate production or in ation.
Q6. What is the scalar series to be forecast?
Let ytC1 denote the scalar series to be forecast and let, Xt be an N -dimensional multiple time series of predictor variables, observed for t D 11 : : : 1 T , where yt and Xt are both taken to have mean 0.
Q7. What is the definition of diffusion index?
Generally speaking, the diffusion index forecasts based on a small number of factors (in most cases, one or two) are found to perform well, with relative performance improving as the horizon increases.