|
Article Excerpt
THE CHARTER OF the Federal Reserve System established 12 geographic Districts with an aim, in part, to provide local influence in monetary policy and to exploit regional information advantages in evaluating aggregate conditions. The latter objective is fulfilled through the collection and dissemination of the Beige Book, an anecdotal summary of regional economic conditions. How well the Beige Book reflects both regional and aggregate conditions is a question that has been studied in a number of papers (e.g., Fettig, Rolnick, and Runkle 1999, Ginther and Zavodny 2001, Balke and Petersen 2002). More recently, some studies have also analyzed how well the Beige Book predicts financial variables such as interest rates and equity prices (e.g., Zavodny and Ginther 2005). Unfortunately, only a weak consensus has been reached on the Beige Book as an indicator of either current or future aggregate economic conditions. Various studies of the value of the Beige Book as a predictor of aggregate conditions have found it to depend on (i) the section of the Beige Book (e.g., the average of regional conditions or the national summary) used as a regressor, (ii) the aggregate measure of economic conditions (e.g., gross domestic product (GDP), employment) being evaluated, and (iii) the timing of the Beige Book release relative to the aggregate data release.
Balke and Petersen (2002), for example, studied Beige Book reports released between July 1983 and January 1997, giving numerical scores to the national summary, national sectoral reports, and each Federal Reserve District report. Because the Beige Book is not released at regular intervals, Balke and Petersen took particular care in specifying the timing of the Beige Book release dates relative to other indicators of economic activity. They found that their Beige Book measures tracked current real GDP growth well. They also concluded that the Beige Book had predictive content for current- and next-quarter real GDP growth beyond that of alternative indicators such as the Blue Chip consensus forecast. Balke and Petersen additionally concluded that the Beige Book appeared to identify turning points sooner than most of the alternative indicators. These results contrast with previous studies (e.g., Fettig, Rolnick, and Runkle 1999), which found that the Beige Book explained a minority of the variation in current-quarter real GDP growth. (1)
Other papers have examined the information content of the regional Beige Book reports to determine their ability to track current regional economic activity. Balke and Yiicel (2000), for example, examined whether the Dallas Fed's Beige Book tracks current Texas real gross state product (GSP) growth and employment growth. Using the Beige Book scores of Balke and Petersen (2002) for the Eleventh Federal Reserve District and quarterly estimates of Texas real GSP growth (Berger and Phillips 1995), Balke and Yticel found that the regional Beige Book indices track the Texas economy well. They also found that the Beige Book contains some information not present in other indicators of economic activity. Ginther and Zavodny (2001) performed a similar exercise using the Atlanta Fed's Beige Book and measures of quarterly regional economic activity in the Sixth District. (2) They found that the Sixth District Beige Book score tracks the regional economy well. However, because the regional and national Beige Book scores are highly correlated, Ginther and Zavodny found that the regional Beige Book provides little additional information when the national Beige Book summary is included in the analysis.
This paper addresses three issues in the current literature. First, because the information contained in the Beige Book is, for the most part, informal and anecdotal in nature, the Beige Book narrative on economic activity is difficult to quantify. The language of individual Districts varies greatly. (3) Previous studies that relied on human readers required careful and precise efforts to minimize the subjective scoring of the Beige Book. In Balke and Petersen (2002), for example, after removing references to calendar years, each author read the reports in random sequence; the resulting index was constructed from an average of the two authors' scores. Thus, even when keeping the same readers, these efforts make replication and extension of the data set virtually impossible. Our solution to this problem is to utilize linguistics software that evaluates the degree of "optimism" and "pessimism" of each Beige Book release (Hart 2000b, 2001). While the primary advantage of this type of software is replicability conditional on a known dictionary, it also removes a large degree of the subjectivity associated with human readers.
Second, we formally address the issue of mixing sampling frequencies. Since the Beige Book is released irregularly, the econometrician is left with the problem of either averaging Beige Book scores, attributing multiple Beige Books to each GDP release, or (arbitrarily) assigning a specific Beige Book score to each GDP release. To solve this problem, we adopt an econometric specification (dubbed the MIDAS model in Ghysels, Santa-Clara, and Valkanov 2004) designed to jointly handle mixed data sampling frequencies and control the number of estimated parameters. This framework allows us to incorporate the information content from multiple Beige Books by weighting each Beige Book score as a function of the elapsed time between Beige Book and data releases. Thus, the model evaluates the effect of a sequence of Beige Books leading up to an economic data release.
Finally, we evaluate the predictive power of each region's Beige Book using District-level employment data. The intent of the Beige Book is to provide a measure of local conditions under the assumption that these local conditions reflect, at least in part, current or future aggregate conditions. However, if the Beige Book fails to reflect even regional conditions, the likelihood that it might predict national conditions should be virtually zero. While a few of the papers discussed above do consider the regional sections of the Beige Book, none of these studies analyzes each of the twelve Fed Districts in a consistent framework.
The remainder of this paper is organized as follows. In Section 1, we outline the empirical model and describe the estimation technique. We discuss the data and summarize the content of the Beige Book, the timing of releases, and the method used to score the Beige Book in Section 2. We then address the construction of the regional economic indicator (District-level employment) used for the disaggregate regressions that follow. In Section 3, we present the results of the estimation using the growth rates in real GDP and aggregate employment. Here, we provide some extensions of the baseline model, including expanding the set of regressors. We present the results from the estimation of District-level employment growth on the relevant section of the Beige Book in Section 4.
1. ECONOMETRIC MODEL
Estimating a time-series model in which a measure of the Beige Book enters as a right-hand-side variable requires modeling data sampled at different frequencies. Measures of output (e.g., GDP) are often sampled quarterly, while employment and industrial production are sampled monthly. The Beige Book, on the other hand, is released at irregular intervals, meaning that varying numbers of Beige Books can be observed in any quarterly GDP cycle and multiple employment reports may be released between Beige Books. This difference in sampling frequency is more than an econometric complication; Balke and Petersen (2002) found that the timing of the Beige Book release is an important factor in determining the Beige Book's predictive power. This suggests a decline in the information content of the Beige Book as a function of increased elapsed time between the Beige Book and data releases.
Unlike other studies in which data are simply time-averaged, we can exploit this variation in frequencies by employing the MIDAS model suggested in Ghysels, Santa-Clara, and Valkanov (2004). MIDAS addresses the difference in sampling frequencies between the left- and right-hand side variables by employing a weighted time aggregation. In this application, the weights are chosen to be functions of the elapsed time between sampled data and an estimated vector of hyperparameters.
Let the growth rate of the economic condition of interest (e.g., employment or GDP) and the Beige Book measure be denoted Y and X, respectively. We use the notation [Y.sub.t] to indicate that variable Y is sampled once between period t - 1 and period t (e.g., once per month in the case of employment and once per quarter in the case of GDP). In contrast, we use the notation [X.sup.(m).sub.t] to indicate that variable X is sampled more frequently, m times in the same period (in our case, weekly). Let k denote the elapsed time (in weeks) between the Beige Book and the end of the data-reporting period, and let [k.sup.max] denote an exogenously set time limit in weeks, such that for k > [k.sup.max], the Beige Book has a negligible predictive effect on [Y.sub.t].
The model can then be written in the following form:
[Y.sub.t] = [alpha] + [beta] [[k.sup.max].summation over (k=0)] [GAMMA] (k, [theta]) [X.sup.(m).sub.t-k/m] + [[epsilon].sub.t], (1)
where [alpha] is an intercept term, [beta] is a slope coefficient, and [[epsilon].sub.t] ~ i.i.d.(0, [[sigma].sup.2]). The lag in [X.sup.(m).sub.t-k/m] is expressed as a fraction of the unit interval between t - 1 and t. That is, if t indicates the monthly frequency and the Beige Book is assumed to be sampled weekly, then m = 4. If, for example, the Beige Book is released two weeks before the end of month t, [X.sup.(4).sub.t-2/4] represents the Beige Book measure as of its release...
|
