On 24th September 2020, the Reserve Bank of India placed on its website a Working Paper titled ‘Inflation Forecast Combinations – The Indian Experience’ under the Reserve Bank of India Working Paper Series. The Paper has been authored by Joice John, Sanjay Singh, and Muneesh Kapur, from RBI though the views expressed were their personal ones.
The web site containing the study is given below:
The intertwined study from the above web contains the following details:
Or putting it as explained by the authors, the paper is organized into five sections. Section II presents a brief review of studies on the inflation forecast combination approach. The analytical framework, data, and sources are presented in Section III. The results are discussed in Section IV, with concluding observations in Section V.
Or as a preamble, I would quote the following from the main report as “extract”
“Accurate, reliable, and unbiased forecasts of inflation are critical for the monetary policy decision-making process, more so for the flexible inflation-targeting central bank. Inflation forecasting is, however, turning challenging across countries.
This paper explores the merits of forecast combination approaches for improving the inflation forecasts in the Indian context. The results seem encouraging.
The inflation forecast combination approach based on the performance-based weighting scheme outperformed the individual models both for headline inflation as well as core inflation for the longer horizons relevant for monetary policy.
Overall, while the performance-based inflation forecast combinations add value to the forecasting exercise, ongoing structural transformations, the greater role of global factors, and recurrent weather shocks continue to pose challenges to the forecasting process.”
Given these complexities of inflation dynamics, central banks often use a suite of models’ approach, supplemented with informed judgment, for improving the quality of the forecasts. Underlying this preference is a tacit recognition that all models are misspecified in some dimension and at some points of time. In this context, a forecast combination approach – combining forecasts from alternative models through a judicious weighting system – finds favor among practitioners.
Against this backdrop, this paper empirically examines the performance of the forecast combination approach for inflation over individual models, the benchmark random walk model, and the median/mean forecasts of inflation from the Survey of Professional Forecasters (SPF) conducted by the RBI.
As regards the individual models, the paper considers 26 individual models and 3 different combination approaches for the period Q1:2001-02 to Q4:2018-19 for the comparative assessment. While headline inflation remains the target for monetary policy, core inflation often provides a better indicator of underlying inflation and the future inflation path (Mishkin, 2007).
Hence, the forecast combination approach has been attempted for core inflation also. The paper’s empirical analysis shows that the performance-based weighting scheme outperforms the individual models both for headline inflation as well as core inflation prices, and exchange rate –
What about foreign central banks’ approach towards inflation forecasting?
Moving specifically to forecast combination approaches in the context of inflation modeling, a number of central banks such as the Bank of England, Sveriges Riksbank, Reserve Bank of New Zealand, and Bank of Canada make use of combination approaches to improve upon the individual model forecasts or to do a cross-check on the forecasts of their key models.
For the UK, while the individual models did not often beat the forecasts from the benchmark autoregressive model, the combination forecasts frequently outperformed the benchmark (Kapetanios, Labhard, and Price, 2008).
In the United States, the results of combination approaches were encouraging.
For India, the combination forecasts were found to be more accurate than
the eight individual models – random walk (RW), autoregressive (AR), moving average with stochastic volatility (MA-SV), vector autoregression (VAR), Bayesian VAR, VAR, and BVAR with exogenous variables (VAR-X and BVAR-X, respectively) and Phillips curve (PC) (RBI, 2017).
Similarly, Dholakia and Kadiyala (2018) considered RW, vector error correction, ARIMA, ARIMA-X, VAR and VAR-X models for the evaluation and found that no individual model outscored others at all horizons.
The exact specifications/ representations of these models are given on pages 9-14 for persons of serious study and understanding.
The following inflation figures from the study are mind-blowing and refreshing for a learning professional.
A brief review of the inflation dynamics since the early 2000s indicates that inflation was rather moderate during 2001-2007; it rose to double-digit levels in 2010 and saw a substantial disinflation from 2014 (Chart 1).
The headline inflation declined to 2.5 percent in Q4:2018-19 from 11.1 percent in Q4:2010-11.
Beginning in 2007, CPI inflation rose mainly due to higher global commodity prices, especially those of crude oil. A deficit monsoon led to a further rise in food inflation in 2009 and its persistence contributed to elevated inflation expectations and generalized inflation.
The double-digit inflation led to a review of the extant multiple indicators framework of monetary policy and a phased switch to a flexible inflation targeting framework in 2014 (RBI, 2014).
This narration leads one to the most important feature of this study that in 2016, the flexible inflation targeting was formally adopted following amendments to the Reserve Bank of India Act, 1934.
A monetary policy committee was constituted, with the objective of achieving the medium-term target for consumer price index (CPI) inflation of 4 percent within a band of +/- 2 percent, while supporting growth. This continues to be the hallmark of monetary policy and the usage of inflation forecasting.
The reforms in the monetary policy framework, a sharp fall in crude oil prices, and better supply management policies contributed to a sustained disinflation from 2014 onwards.
In view of recurrent food-related shocks, this period also witnessed episodes of divergence between headline inflation and core inflation measured by excluding food and fuel (Raj et al., 2020). The large swings in the inflation dynamics over the past decade clearly point to the forecasting challenges.
Chart 1: CPIC – Year-on-year Inflation – Headline and Core from June 01 to December 18 are projected on page 15.
An evaluation of the RBI’s inflation projections indicates that the forecast errors were comparable to other countries. The modeling and forecasting approaches are constantly reviewed and refined by staff, and information collecting systems strengthened on an ongoing basis to minimize forecast errors (Raj et al., 2019).
Like other central banks, the RBI regularly conducts a survey of professional forecasters (SPF) wherein different professional forecasters give their individual projections for a set of macroeconomic indicators, including inflation. The results of the survey are published by the RBI in terms of mean and median of the individual forecasts – these results can, therefore, be interpreted as a variant of the simple average combination approach being studied in this paper.
The headline inflation measure is based on the consumer price index (combined) (CPIC). Core inflation is often calculated by removing the volatile components/ sub-groups in the consumers’ consumption basket.
Although there is no official measure of core inflation, CPI excluding food and fuel in the Indian context is often treated as a suitable measure of core inflation (Raj et al., 2020). Hence, CPIC inflation excluding food, fuel and light is taken as the measure of core inflation.
The National Statistical Office (NSO) started compiling CPIC in 2011; RBI (2014) provided back-casted data on CPIC for 2001-2010, using data on CPI-Industrial Workers (CPI-IW)6. Therefore, the period of study was taken from Q1:2001-02 to Q4: 2018-19 and data frequency was chosen as quarterly7.
As the CPIC prior to 2011 in RBI (2014) was back-casted largely based on the retail prices faced by industrial workers, the paper also undertakes, as a robustness exercise, analysis for the smaller sample period (Q1:2011-12 to Q4:2018-19) for which the actual data on CPIC are available.
Analysis of data by charts given in the paper is given below for serious study.
Chart No 1. Chart 1: CPIC – Year-on-year Inflation – Headline and Core
Chart 2. Headline Inflation Forecasts (Full Sample: Q1:2001-02 to Q4:2018-19)
Chart 3: Headline Inflation Forecasting Performance: RMSEs Relative to RW Model for Best Individual Model and Forecast Combinations: (Full Sample: Q1:2001-02 to Q4:2018-19.
Chart 4: Core Inflation Forecasts (Full Sample: Q1:2001-02 to Q4:2018-19)
Chart 5: Core Inflation Forecasting Performance: RMSEs Relative to RW Model for Best Individual Model and Forecast Combinations (Full Sample: Q1:2001-02 to Q4:2018-19)
Chart 6: Out-of-Sample RMSE of Inflation Forecasts: Model-based Forecast Combination and Professional Forecasters.
My review of the charts indicated that CPIC headline inflation showed a maximum of 12% for the quarter ended in March 2010.
What are the conclusions as per the study of inflation forecasting after a deep study of various factors over a massive period of 2001-2019? Yes, I repeat that our young professionals surpass the best in the world by their accurate calculation techniques and deep application for practical results.
Inflation forecasts are the key inputs for monetary policy formulation by inflation-targeting central banks. Inflation forecasting has become a more challenging task due to the weakening of the traditional link between inflation and economic activity across countries for a variety of factors such as greater external openness, volatile exchange rates, and commodity prices, increased competition from e-commerce, and potential non-linearities.
In this milieu, a forecast combination approach – combining forecasts from alternative models through a judicious performance-based weighting system – can potentially enhance the forecasting performance of the individual models.
This paper empirically examined the forecasting performance of the combination approaches in the Indian context relative to a wide range of individual models spanning different modeling frameworks. Although the combination forecasts significantly improve upon the individual models, the absolute forecast errors of the combination models are non-negligible. A part of these errors is due to the large recurrent fluctuations in the key conditioning variables such as crude oil prices and exchange rate movements. Large shocks from the food side also contribute to the forecast errors.
The empirical analysis showed that even the simple average of the forecasts based on individual models was comparable with the ‘best’ performing individual model’s forecast. The performance-based weighting schemes outperformed the individual models both for headline inflation as well as core inflation by a substantial margin at the longer horizons.
The basic purpose of this article is to invite the attention of all professionals to consider the tremendous industrial research combined with the best econometric techniques interspersed with the most advanced mathematical models totally based on available economic data for inflation forecasting undertaken by the most brilliant economists from RBI and appreciate the efforts to understand our economy which has merged with the world economy. The tremors of the Indian economy attract maximum investment from the rest of the world.
Yes, I confirm that the day is not far when we would be the No. 1 economy of the world.
Disclaimer: Having been attracted by a brilliant study published by RBI on its website, I collected many of the ideas which appealed to me. Many ideas that are too much theoretical have been reproduced since I could not explain them further. Neither taxguru.in nor RBI is responsible for my views. Anyone serious to understand is requested to go to the RBI website and read 32 pages research study for keener understanding and application.