Recent observations by some policymakers that government policies are premised on wrong inflation forecasts have brought back into the limelight discussion over the right measure of inflation and the method of its forecasting. Many argue that the current practice of year-on-year estimation of inflation numbers might not reflect the true built-up in price pressures in the most recent period. Discussions in recent times have also been about the way indicators used by policymakers are presented and interpreted, mostly with respect to inflation estimates reported by RBI. The argument is that the seasonally adjusted month-on-month annualised rates will give a better picture of recent developments (built-up in prices in case of inflation) in the economy as compared to the year-on-year growth rates. It is also pointed out that this the standard practice among most of the industrialised nations.

With relation to the relative merits of alternative ways of estimating various growth rates (inflation, in particular), the two rates may give completely different pictures of economic developments because of the methodology adopted. What emerges is that when data is de-seasonalised, the information content from such estimates provides for predictability and exposition. There are other issues such as uniformity of predictors that have a close relation with inflation estimates and the choice of seasonal adjustment. The suggested and most widely used method is X-12-ARIMA, which is made available by the US Census Bureau for seasonally adjusting the data, but which may not be consistent with seasonality in Indian data.

Major conflicting areas in seasonal adjustment are precision, user-friendliness and the reliability of methodology. In a broad sense, seasonal adjustment means removing calendar variations and trading day effects. Calendar variation?including holiday effects,

regular policy announcements, weather and season effects, etc?is mostly ignored in Indian data. But as activities like holidays or festivals here are governed by a national calendar that is not in line with the standard international calendar followed in international markets, automatic application of the international method for Indian data might lead to spurious conclusions.

The existing procedure of X-12-ARIMA is designed to give seasonal adjustments where the sum of adjusted and raw series is the same for a calendar year. It also needs the full set of data for a calendar year to update the seasonal estimates. Hence, for example, if we have

data for January 2012, the method would not update the seasonal component?rather it takes January 2011 seasonality information. Thus, the seasonal factors obtained may not be precise and stable due to the limitation of X-12-ARIMA.

Despite the above limitations, if we want to de-seasonalise, then the indirect seasonal adjustment or seasonally adjusting the components before aggregating them usually gives a better seasonally adjusted series, where the component series have quite distinct seasonal patterns. Given that the different components of the aggregated WPI data?food, fuel and lubricants, primary articles, etc?exhibit different patterns of seasonality, the seasonal pattern for the aggregate WPI can be better estimated by seasonally adjusting each component separately and then summing up to get the final aggregated and seasonally adjusted series as per the weighting scheme.

Once the issues related to the seasonal noises are addressed, there comes the important question concerning which series should be made public and tracked by policy analysts. Apart from the raw and the seasonally adjusted series, the other potential candidates are year-on-year changes, series annualised by compounding, and the period-to-period annualised growth rates. All these alternatives need a deeper analysis with respect to the message they convey, because the extent of variation varies for different growth rate calculations while using the sub-annual series. For instance, for the period April 1994 to September 2011, we see that the year-on-year changes are less volatile compared to the other two calculated series, with standard deviations of 3.29, 8.69 and 8.04, respectively, for India. This volatility trend is not different in advanced economies as the standard deviations are 1.14 and 3.48 for year-on-year and month-on-month, respectively, for the US. Volatility and predictability are inversely related. In general, price formation (using year-on-year) in India is less predictable compared to the US, largely due to the presence of distorted markets. Within India, we may note that alternative procedures end up showing much higher volatility compared to year-on-year, thus making the series less predictable.

In sum, while de-seasonalised series are important, given the limitations of the existing methods, the government needs to release only the raw data and the issue of de-seasonalisation needs to be left to the judgement of the researchers. As for policy, given that policy analysis should be done using a less volatile, compatible and more predictable series, the usual practice of year-on-year changes remains more appealing and informative despite its limitations. Most of the time the macro policy is based on medium- to long-term growth cycles rather than short-term fluctuations, and cyclical patterns are much more clear in year-on-year estimates than in month-on-month. Further, although it is necessary to observe recent fluctuations, given the structure of market and distorted information coming out of it, this restricts the use of suggested alternatives to year-on-year estimates. In India, substantial revisions in most of the macro information also makes month-on-month estimates less useful for predictability and, hence, policy formulations based on such estimates.

The authors are with the National Institute of Public Finance and Policy, New Delhi. Views are personal