داده تجدید نظر شده در هند: پیامدهایی برای سیاست های پولی

This paper studies data revision properties of GDP growth and inflation for the Indian economy. The results show that revisions to GDP growth and inflation are significant, and cannot be characterized as either containing pure news or pure noise. We also find that there is a significant predictable component in the revisions to GDP growth and inflation. Our findings suggest that if the Reserve Bank of India were to follow a Taylor rule for its monetary policy formulation, then the interest rate based on the preliminary data would be much lower than the one based on the fully revised data.

Most macroeconomic time-series data are subject to revisions. Data revisions pose a problem for policymakers since they formulate policies in real-time without access to the fully revised data. Data revisions also create a challenge for economic researchers since their empirical work is based on heavily revised data, and the policy conclusions based on the use of heavily revised data can often be misleading in real-time. Recently there has been a surge in literature on data revisions and the implications of these revisions for policy making.1 Most of the research on data revisions has focused on OECD countries, especially on the U.S. economy. The issue of data revisions in developing countries has not attracted a great deal of attention due to non-availability of real-time data.2 The macroeconomic time-series data in India are also subject to revisions. GDP growth and the Wholesale Price Index (WPI) – the primary measure of inflation in India – undergo heavy revisions. Fig. 1 and Fig. 2 plot revisions to GDP growth and aggregate inflation in India. It is evident from the graph that the final estimate of inflation is higher than the preliminary estimate for the majority of the sample since revisions are almost always positive. The revisions to GDP growth show higher volatility before 2001, and has been mostly positive, though less volatile, since then.This paper studies the data revision properties of GDP growth and the WPI inflation and its sub-components in India, namely primary, manufacturing, and fuel inflation.3 We examine whether data revisions to GDP growth and inflation have zero mean, and whether they can be forecasted using the information available at the time of the preliminary data announcement. This fits in with the news versus noise literature in the data revision that has been studied extensively for the key macroeconomic variables in the U.S.4 Conventional wisdom also suggests that data revisions pose a serious problem for monetary policy formulation, as the monetary policymakers are uncertain about the true state of the economy on the basis of preliminary estimate of macroeconomic variables. Thus, we examine the effect of data revisions on monetary policy formulation in India. Specifically, we examine how the interest rate prescribed by a Taylor rule would differ if the real-time data were used instead of the fully revised data. Literature on data revisions has become quite extensive after compilation of the real-time data set at the Federal Reserve Bank of Philadelphia by Croushore and Stark (2001). Croushore and Stark (2001) show that data revisions pose a serious challenge for policy formulation as well as the econometric estimation of macroeconomic models. A big part of the literature on data revisions investigates the revision properties, and tests whether these revisions are predictable or not. Mankiw, Runkle, and Shapiro (1984) tested whether preliminary announcements of the money stock were rational forecasts of final announcements. Mankiw and Shapiro (1986) applied a similar analysis to GNP data. Faust, Rogers, and Wright (2005) tested the news versus noise hypothesis for revisions to the OECD output data, and found evidence in support of the noise hypothesis. Croushore (2008) studied the patterns of data revisions to the inflation rate in the U.S., and found that it is possible to forecast revisions from the initial release. He noted that the initial release of inflation is likely to be revised up. We find that revisions to GDP growth in India between 1997 and 2001 are characterized by two regimes: volatile and insignificant revisions between 1997:Q2 and 2001:Q1, and mostly positive and significant revisions after 2001:Q1. The revisions to GDP growth after 2001 are also associated with lower volatility. Data revisions to inflation in India are always significant. Specifically, we find that revisions to the WPI inflation and its sub-components are likely to be revised up, as the preliminary estimates are too low. This effect is especially pronounced for manufacturing inflation which accounts for the biggest share of aggregate inflation. Our results show that the data revisions to output growth and inflation in India cannot be strictly characterized as either containing pure news or pure noise. There is evidence of predictability in data revisions using the preliminary announcement. We find that 39% of the variations in the data revisions to GDP growth between 1997:Q2 and 2001:Q1 can be explained using the preliminary release of GDP growth, whereas the corresponding explanatory power of the initial release is 22% after 2001:Q1. Around 17% of the variations in data revisions to aggregate inflation can be explained by the initial release of the inflation. The degree of predictability is greater for revisions to the manufacturing component of inflation as compared to the fuel and light and the primary products. Preliminary estimates of manufacturing inflation explain 39% of the variations in revisions to manufacturing inflation. The greater degree of predictability for the manufacturing inflation is consistent with the late arrival of source data for the manufacturing component of the WPI. The rejection of the pure news and the pure noise hypothesis is consistent with what has been found by Mork (1987), and Aruoba (2008) for most of the U.S. macroeconomic time series data. The ex post predictability of data revisions does not imply that data revisions are forecastable in real-time. To investigate the real time predictability of data revisions to inflation, we compare the forecast error of revision forecast generated in real-time with the actual revision (assuming preliminary data as forecast of the final data), and we find that revision forecasts can be substantially improved upon in real-time. Our results show that ignoring data revisions can have serious implications for monetary policy formulation. We find that if the Reserve Bank of India were to follow the Taylor rule in setting the interest rate, then the interest rate based on heavily revised data is on average higher than the one based on the real-time data. The Taylor interest rates based on the preliminary data tends to be usually too low. This implies that if the Reserve Bank of India does not take into account data revisions that could take place in the future, then the monetary policy action may turn out to be too expansionary ex post. The plan of the remainder of this paper is as follows: Section 2 discusses the data and its properties. Section 3 studies the properties of data revisions in India. Section 4 deals with forecasting of data revision process. Section 5 assesses the impact of data revisions on monetary policy formulation in India and Section 6 summarizes the main results.

This paper studies data revision properties of GDP growth and inflation, and their implications for monetary policy formulation in the Indian economy. While the GDP undergoes multiple rounds of revisions, WPI, which is the primary measure of prices in India, and is used widely for policy deliberations undergoes one round of revision with a lag of eight weeks. We find that revisions to WPI inflation and its sub-components are positive and significant. The results indicate that on average the final estimate of inflation is higher than the preliminary estimate, and is likely to be revised up. We find that the revisions to GDP growth were volatile and insignificant before 2001, but were positive and significant after 2001. Our results show that the data revisions to GDP growth and WPI inflation and its sub-components cannot be characterized as either containing pure news or pure noise. The optimal use of data revision properties can be used to bridge the gap between the final and the preliminary estimate. Our findings suggest that around 22% of the variations in revisions to output growth after 2001:Q1 can be explained using the preliminary estimate of GDP growth, whereas the corresponding number for aggregate inflation is 17%. The degree of predictability for data revisions to manufacturing inflation is higher than other sub-components of inflation. We find that 39% of variations in revisions to manufacturing inflation can be explained using the first release. The results obtained in this paper suggest that ignoring data revisions to GDP growth and WPI inflation can have significant policy consequences. If the Reserve Bank of India were to follow a Taylor rule in monetary policy formulation, then our results indicate that monetary policy based on the preliminary data may prove to be too expansionary ex post.