- The methodology of the construction of composite index for Indian Manufacturing output are summarized as follows:
- Selection of the reference series that represents the manufacturing activity
- Macroeconomic variables that contain information on reference series (potential leading indicators)
- Develop multiple leading indicators and evaluate their performances in sample and out of sample
- Select the leading indicator showing the best performance.
Selection of the reference series that represents the manufacturing activity
- The very first step in the leading indicator approach is to choose a proxy for the manufacturing activity. Gross Domestic Product (GDP) and Index of Industrial Production (IIP) are generally used as measures for economic activity. The GDP is available at quarterly frequency and it is published with a lag of more than a quarter. The number of points available for the recent time period will be less as the reporting cycle is large. This leads us going back a long duration in the past. To gain a decent sample size to construct leading indicator models we need to go very long in the history. However, it may not be good to go long back in the history as Indian economy has seen structural changes over the past two decades.
- Subsequently, for the leading indicator approach, a series that is available at a high frequency and published with least possible delay is preferred as the reference series. For this reason, in this study, IIP has been chosen as reference series and has the advantage of being a monthly reported variable reported at a lag of 4 to 6 weeks.
Macroeconomic variables that contain information on reference series (potential leading indicators)
- A large data set was constructed to cover the variables that represent all economic spheres of activities. We have chosen economic series that is easily and quickly available, and the series should not be subject to major revisions else that would impact the earlier conclusions with the publication date of the series being timely. Most importantly there should be an economic rational that the series precede the refer- ence series. We initially shortlisted around 70 different variables for the analysis.
- The shortlisted data series included various thematic stock market returns, various IIP components such as mining activity, electricity production, SBI lending to various sectors, NPAs, exchange rates, interest rates, global economic growth rate, commodity prices. We have considered futures and spot prices where available. SBI lending data has provided significant insights into the future economic activity in industry and confidence within industry.
- We have created two indices namely SBI Monthly Composite Index and SBI Yearly Composite Index. Both the indices fulfill complementary purposes such as month on month sentiment movement vs. year on year growth forecast respectively. The complementary nature of these indices help user to gauge the sentiments in the markets. If a month on month forecast is consistently negative, it can lead to negative growth rate in year on year index after a while. A positive growth rate of year on year growth index may fuel the growth momentum and increase the sentiments for monthly prediction.
- We have constructed leading indicators using various statistical methods, including simple regression based models (Linear regression models, logistic models, factor models) and VAR based models suggested by Stock and Watson. After choosing the best possible models on statistical parameters, we have tested the models on in sample forecasts and out of sample forecasts.
- Leading indicators are evaluated for directional predictions out of sample at turning points and compared with performance of other leading indicators such as PMI.
To empirically evaluate the lead lag relation for each identified variables we have conducted correlation analysis at various
lags with the reference series. We also evaluated the lead times in turning points between possible leading indicator and reference series. To weed out spurious variables, we conducted Granger causality for the variables.
- After identifying the potential leading indicators, we constructed models to explain the variance in the monthly growth in IIP series. After thorough statistical analysis we finalized on three models, All the three forecasts had an adjusted R square greater than 0.75.
- We evaluated the performance of above models on in-sample and out of sample forecasts on parameters such as Root Mean Squared Error (RMSE), Mean Absolute Deviation (MAD), on percentage directional accuracy.
Chart 2 provides SBI Yearly Composite Index performance over a large period of 10 years, it clearly depicts that SBI Composite Index has closed followed the trend and variance in the IIP Manufacturing index.
As per the above compiled results for the most recent years beginning Apr’13
- In sample prediction for SBI Composite Index during Oct’06-Mar’13 is at 71% (PMI at 49%).
- Also, the out sample prediction during Apr’13-Sep’14 is at 72% (PMI at 50%).
- We foresee a stronger manufacturing recovery in Dec’14.