RBI Working Paper Series No. 02
Tail Risks of Inflation in India
Silu Muduli and Himani Shekhar
Abstract
This paper estimates the tail risks of consumer price inflation in India using a quantile regression framework. It examines the impact of various domestic and global macroeconomic factors, along with the role of the flexible inflation targeting (FIT) in influencing the inflation tail risks as well as in explaining the shifts in its conditional distribution. A rise in domestic income and household inflation expectations, elevated global commodity prices – both fuel (i.e., crude oil) and non-fuel, and easy financial conditions pose upside risks to Consumer Price Index (CPI) headline inflation. The results also show that both lower and upper tail risks of inflation have stabilised in the FIT period and that the macroeconomic factors capture the tail risks to the inflation target band of 2 to 6 per cent in India reasonably well.
JEL Classification: C21, C53, E31, P24, P44
Keywords: Inflation at risk, inflation uncertainty, monetary policy
Introduction
A precise estimate of the future inflation path and uncertainties around it is crucial for a proper assessment of inflationary conditions. During extreme events, such as the global financial crisis (GFC) and the COVID-19 pandemic, it becomes very difficult to predict the trajectory of inflation due to uncertainties surrounding it. In such circumstances, the distribution of future inflation, in addition to the inflation forecast, may be useful for future guidance, particularly under a flexible inflation targeting (FIT) framework.
For India, food – a major component of the consumer price index (CPI) basket – typically has higher volatility owing to supply-side issues and the monsoon dependence of Indian agriculture. As a result, any disruption in weather patterns gets reflected in production, and in turn, prices. Extreme weather events namely, excess rains, deficient rains, floods, cyclones etc., bring additional uncertainty around the conditional mean trajectory of food inflation and makes it less reliable. Similarly, the high dependence of India on crude oil imports makes it susceptible to any global oil price shock.
The uncertainty around food price inflation spills over to the uncertainty around CPI headline inflation as food occupies a significant proportion of the CPI basket and is also susceptible to many supply shocks. In such times of uncertainty, along with the mean path of future inflation, the distribution of inflation becomes crucial in the assessment of inflationary conditions. The inflation distribution may help in a proper assessment of the upside and downside risks to inflation.
The upside and downside risks to inflation are known as Inflation at Risk (in notation, IaR) in the literature, a measure similar in notion and concept to Value at Risk (VaR) in financial risk-management theory to estimate the market and credit risk of a portfolio (Andrade, Ghysels, and Idier, 2012; Banerjee et al., 2020; López-Salido and Loria, 2020). The conventional approach assumes the symmetric distribution of errors around the mean path, which, however, may not always hold. Thus, the asymmetric nature of future inflation distribution may be useful in explaining the tail risks (i.e., the possibility of extreme values on either side) of inflation and in helping the monetary policy in communicating the balance of risks.
Besides the asymmetry, understanding and accounting for the uncertainty around the central tendency are also crucial in stabilising inflation. Inflation uncertainty is one of the primary costs of inflation to the real economy as expected inflation is an important factor while making economic decisions. Uncertainty surrounding future inflation creates uncertainty regarding the future value of savings and investments, which in turn, may distort the efficient allocation of resources (Chowdhury, 2014). Consequently, inflation uncertainty can adversely affect consumption, investment, and growth. Since the monetary authority’s primary objective is to stabilise prices, it is also crucial to empirically examine whether it accounts for inflation uncertainty in monetary policy formulation.
Given the importance of the distributional characteristics of inflation, the paper derives a conditional distribution of inflation which also indicates the balance of upside or downside risks. The paper has the following objectives:
- Estimate tail risks of inflation in India corresponding to various domestic and global drivers and test whether inflation risks have stabilised in the IT period;
- Examine shifting of the conditional distribution of inflation for different domestic and global shocks;
- Estimate expected tail values of inflation and examine the robustness of the models;
- Examine the causal relationship between inflation and inflation uncertainty and assess the reaction of the monetary policy to the asymmetric nature of conditional distribution and uncertainty around the central tendency.
Estimations of conditional distribution and tail risks are based on a hybrid version of the standard New Keynesian Phillips Curve (NKPC) framework with financial conditions, crude oil price, and exchange rate of the Indian rupee (INR) vis-a-vis US dollar (USD) and global demand conditions- proxied by US real GDP growth as additional explanatory factors (Auer, Borio, and Filardo, 2017). The paper also examines the role of the IT framework in stabilising the tail risks. Therefore, the study includes both demand and supply-side factors in the estimation of the conditional distribution. In the Indian context, the existence of the Phillips curve has been established based on samples at national as well as sub-national levels in the post-GFC period (2007-09) (Behera, Wahi, and Kapur, 2018; Salunkhe and Patnaik, 2019). In this regard, a recent study by López-Salido and Loria (2020) concluded that tail risks are sensitive to domestic economic slack and have a significant role in influencing inflation distribution. Some recent studies have also considered the financial conditions index to examine the upside and downside risks to inflation during tight/easy financial conditions (Chevalier and Scharfstein, 1996; Gilchrist, Schoenle, Sim, and Zakrajšek, 2017). Chevalier and Scharfstein (1996) argue that during tight financial conditions, firms that face a higher constraint on accessing liquidity may set a higher price to increase their cash flow leading to higher inflation.
India adopted the flexible IT framework in 2016, which was reviewed in March 2021, wherein the inflation target of 4 per cent with a ±2 per cent band around it was continued until the next review in 2026. During the period from 2011-12 to 2013-14, the average CPI-C inflation rate stood at 9.4 per cent, which moderated gradually towards the midpoint of the target band during the IT period. Notably, CPI-C inflation averaged 3.9 per cent during the IT phase of October 2016 – March 2020. RBI (2021) highlights the success of the framework in anchoring inflation expectations of both households and professional forecasters and lowering average inflation.
Given the relevance of inflation projections in policy formulations, the RBI regularly publishes a fan chart of asymmetric inflation distribution based on a two-piece normal distribution which consists of two normal distributions below and above the mean (Banerjee and Das, 2011). Both the pieces have the same mean, but different standard deviations. This difference in standard deviations brings the asymmetry of a distribution around the mean. In the two-piece normal distribution, the values of standard deviations are derived from the past deviations from the forecast values, which incorporates asymmetry in the distribution. However, this paper derives the conditional distribution of inflation based on the estimated quantiles from quantile regression conditioned on the macroeconomic environment rather than depending on the past. Few recent studies in advanced economy central banks have highlighted the usefulness of conditional quantile regression in deriving the fan chart by incorporating expert judgements (for instance, see Sokol, 2021). Therefore, deriving the conditional distribution of inflation based on the current macroeconomic situation rather than depending on the past as in the case of a fan chart is the major contribution of this paper.
The paper analyses the historical tail risks and shifts in the conditional distribution of inflation for various shocks. It concludes that a rise in domestic income and household inflation expectations increases the upside risks and lowers the downside risks to inflation. Elevated global commodity prices of both fuel (i.e., crude oil) and non-fuel, global economic growth and easy financial conditions raise the upside risks to inflation. Further, the results add to the success story of the adoption of the IT framework in India in stabilising CPI headline inflation as both lower and upper tail risks of inflation have stabilised in the IT period. Regarding the predictive efficiency of the models, the paper concludes that the models based on various macroeconomic factors capture the tail risks to the inflation target band of 2 to 6 per cent in India to a considerable extent. While examining the response of the monetary policy to asymmetry and uncertainty of inflation distribution, the paper finds evidence of tightening of the monetary policy during the periods of higher inflation uncertainty.
The rest of the paper is organised as follows. Section II discusses the literature on the conditional distribution and tail risks associated with CPI-C inflation in India. Building on the literature and existing data sets, Section III presents a few stylised facts for inflation in India during the sample period and discusses the methodology used in the paper. Section IV presents the empirical results and also examines the asymmetry and uncertainty pertaining to inflation distribution and the response of the monetary policy to these. The last section concludes the paper with a few policy implications.
Studies that focus explicitly on the analysis of tail risks of inflation are relatively new in the literature (Banerjee, Contreras, et al., 2020; López-Salido and Loria, 2020). However, a few studies in the last decade analyse the quantiles of inflation and their dynamics; for instance, Wolters and Tillmann (2015) examine the persistence of different quantiles in the conditional distribution of inflation. Gupta, Jooste, and Ranjbar (2017) find higher persistence of inflation for higher quantiles in the case of South Africa. In related literature on convergence, Tsong and Lee (2011) in a study of 12 OECD countries find asymmetric convergence of inflation to long-run value post negative and positive shocks using the quantile regression approach. They find that positive shocks are more persistent than negative shocks and converge slowly to the long-run level. Similar empirical evidence is also seen in Uganda (Anguyo, Gupta, and Kotzé, 2020).
Several studies have focussed on various determinants of inflation in a quantile regression framework; for instance, Iddrisu and Alagidede (2020) explain food inflation and its various determinants, such as economic growth, world food price inflation, monetary policy, etc., using quantile regression for South Africa. Lahiani (2019) explores the transmission of crude oil prices to different quantiles of overall prices for the US.
Further, a plethora of studies have focused on the important determinants of conditional mean inflation using Phillips-curve specification and its different versions. Many of these studies find that the Phillips curve relationship estimated at the mean level weakened after the GFC (Blanchard et al., 2015). A few studies extend the NKPC estimation by incorporating global economic slack along with domestic economic slack to explain the domestic inflation dynamics (Auer, Borio, and Filardo, 2017). However, the evidence of global economic slack is mixed. For instance, Forbes (2019) finds significant evidence of the role of global economic slack on domestic inflation, while Mikolajun and Lodge (2016) find a limited impact of it on domestic inflation. Xu, Niu, Jiang, and Huang (2015) use non-linear quantile regression to estimate the Phillips curve for the US and conclude that the shape of the Phillips curve differs across quantiles. In the case of India, researchers find evidence for the existence of the Phillips curve based on samples from national and sub-national data in the post-GFC period (2007-09) (Behera, Wahi, and Kapur, 2018; Salunkhe and Patnaik, 2019).
The above studies focus on exploring quantiles of inflation and do not explicitly mention the tail risks of inflation. Andrade et al. (2012) introduced an explicit analysis of tail risks of inflation i.e., “Inflation at Risk (IaR)” and used asymmetric property and distributional uncertainty to explain the monetary policy rule. The tail risks of inflation have been a major concern, particularly in advanced economies, in the post-GFC period (López-Salido and Loria, 2020).
Some recent studies have also considered the financial conditions index to examine the upside and downside risks2 to inflation during tight/easy financial conditions (Chevalier and Scharfstein, 1996; Gilchrist, Schoenle, Sim, and Zakrajšek, 2017). Firms facing higher constraints on accessing liquidity during tight financial conditions and firms with weak balance sheets may set a higher price leading to higher inflation. In a recent study, Banerjee et al. (2020) use the volatility of asset prices as a proxy for tight financial conditions and find a significant impact of financial conditions on inflation. Some earlier studies based on quantile regression also find a positive association between stock market return and different quantiles of inflation in G7 countries (Alagidede and Panagiotidis, 2012).
Inflation at Risk is similar to the concept of Value at Risk (VaR) in financial risk management, i.e., the extreme quantiles for a given level of probability. Among macroeconomic variables, a similar measure is also available for economic growth as Growth at Risk (Prasad et al., 2019). López-Salido and Loria (2020) in their study of advanced economies explicitly derived the conditional distribution based on quantiles. Banerjee et al. (2020) extended the analysis by including emerging economies and estimated the conditional density. Their study concludes that countries with IT show a relative moderation in inflation risks than non-IT countries. Banerjee, Mehrotra, and Zampolli (2020) model the impact of the COVID-19 pandemic and find higher upside and downside risks to inflation in emerging economies, and higher downside risks in the case of advanced economies.
The advancement in deriving the conditional distribution based on quantile is a novel contribution in the above studies as compared to earlier studies that used quantile regression to study the quantiles of inflation and their dynamics. These studies highlight, in particular, the impact of different shocks on the tail risks of inflation. In our analysis, we augment the above-discussed analysis for India and examine certain country-specific features to explain the dynamics of tail risks of inflation.
In related literature, Andrade et al. (2012) examine the response of the monetary policy to inflation asymmetry and uncertainty based on the professional forecasters’ survey data on inflation. A few studies explicitly analyse the relevance of inflation asymmetry on monetary policy formulation (Andrade et al., 2012; Evans, Fisher, Gourio, and Krane, 2016). Further, there exist several studies that discuss the causality and reverse causality between the level of inflation and inflation uncertainty (Cukierman and Meltzer, 1986; Sharaf, 2015; Su, Yu, Chang, and Li, 2017).
According to Friedman (1977), higher average inflation causes uncertainty about future monetary policy responses, resulting in broad differences in actual and expected inflation, and therefore, leads to economic inefficiency making it detrimental to growth. This relationship was later formalised in a game-theoretic framework by Ball (1992) and the work is known as the Friedman-Ball hypothesis. On the contrary, Pourgerami and Maskus (1987) argue that inflation and inflation uncertainty have a negative relationship and reject the hypothesis of the deleterious effect of high inflation on price predictability as elevated inflation levels lead to the deployment of additional resources for lowering projection error resulting in better forecasts and hence reduction in inflation uncertainty.
Coming to another dimension of a causal link from inflation uncertainty to inflation, Cukierman and Meltzer (1986) postulate that increased inflation uncertainty causes inflation to increase – the Cukierman-Meltzer hypothesis. When policymakers act with low credibility, the ambiguity of goals and poor quality of monetary control, then it may lead to an increase in the average inflation rate (Rojas, 2019). The flip side of this hypothesis is proposed by Holland (1995) who concludes that higher volatility of inflation reduces price levels reflecting policy makers’ motives for stabilisation. Further, sometimes the bidirectional relationship between inflation and inflation uncertainty is also observed under the Friedman-Ball hypothesis and the Cukierman-Meltzer hypothesis – higher inflation will increase the inflation uncertainty and vice-versa. In the case of India, Chowdhury (2014) finds evidence in support of this hypothesis using the generalised autoregressive conditional heteroscedasticity (GARCH) model. Kundu, Bhoi, and Kishore (2018) also present similar evidence between inflation and inflation volatility graphically at the sub-national level.
Inflation uncertainty is a major concern for the monetary authority, as it assigns weights inter-temporarily to minimise its loss preference. Although not explicitly, a few studies have shown the detrimental effect of high inflation uncertainty on economic activity (Sauer and Bohara, 1995; Zhang, 2010). Hence, it becomes imperative for the monetary authority to reduce inflation uncertainty through appropriate policy instruments (Gan, Yee, Hadi, and Jalil, 2019; Zhang, 2010). In a standard monetary policy rule, besides output gap and inflation, studies have included exchange rate, global policy rate, and global economic growth to examine their impact on the monetary policy rate (Hutchison, Sengupta, and Singh, 2010; Reserve Bank of India, 2021). However, there are very few studies that explicitly model inflation uncertainty in the monetary policy rule to estimate its impact (Andrade et al., 2012).
This paper adds to the limited literature in the Indian context on the importance of tail risks of inflation and their role in monetary policy. In this paper, we derive the conditional distribution of inflation based on the quantile regression in an NKPC framework that incorporates the macroeconomic environment. Further, we examine the role of distributional asymmetry and uncertainty in the monetary policy formulation. Our results support the causality between inflation and inflation uncertainty.
III.1. Stylised Facts
The uncertainty around food price inflation spills over to the uncertainty around headline inflation as food is a major contributor to CPI headline inflation variance (Chart 1)3. Within the FIT framework, price stability – avoiding high inflation rates or very low inflation rates over time – is the primary mandate for the RBI as volatile prices distort the economy’s price signals and may result in the misallocation of resources.
The CPI-C headline inflation has undergone significant changes in its distribution over the last decade in line with the evolving macroeconomic conditions (Chart 2a). Too high or too low inflation representing the upside and downside tail risks to inflation, respectively, is detrimental to RBI’s secondary objective of growth as well, and thus, requires a proper assessment of these risks. The FIT period coincided with a moderation in CPI inflation on the back of consecutive years of bumper food grains and horticulture production and relatively stable global commodity prices, particularly the crude oil. Irrespective of the broad easing of inflation, headline inflation deviated from the target, and went once below the lower bound of 2 per cent (in June 2017) and above the upper bound of 6 per cent consistently during December 2019-December 2020 mirroring the developments in food prices primarily owing to monsoon-related shocks and the COVID-19 pandemic-related supply disruptions (Chart 2b).
Given the high weight of food in the CPI basket, which is susceptible to adverse supply shocks, pressures in the food basket not only drives up the CPI headline inflation but also contributes substantially to its variance (Chart 3a and Chart 1). High levels of CPI inflation are often accompanied by higher volatility (Chart 3b). High inflation rates accentuate concerns about future inflation as they have the potential to influence long-term interest rates and generate uncertainty about the future value of the investment, savings, wages, tax rates, etc. In light of the fact that price stability is the primary objective of central banks, the volatility around the inflation path warrants the attention of the monetary authority. The asymmetry of distribution assumes greater importance given the fact that economic agents’ expectations are influenced by the distribution of the realised inflation (RBI, 2021).
III.2. Data and Summary Statistics
This paper uses the monthly CPI-C inflation rate (y-o-y, per cent) from September 2009 to December 2019 (before 2011, CPI-IW (CPI Industrial Workers) is used). Table 1 presents the summary statistics of variables used in the paper. The CPI-C inflation averaged around 7 per cent during the sample. It remained above average during 2010-2012. It moderated subsequently to around 4 per cent with RBI’s formal adoption of the FIT framework in 2016, which was preceded by a transitional glide path from 2014-15. The one-year ahead median inflation expectations of households have been used as a proxy for a forward-looking measure of inflation expectations4 which averaged around 11.33 per cent during the sample. As households’ inflation expectations are observed quarterly, a cubic spline methodology has been employed to convert it into a monthly series (Stuart, 2018). Given the fact that the inflation expectations series is relatively persistent and stable, the above methodology might be a better approximation to obtain monthly frequency data.
GDP growth at the constant market price has been used as a proxy for demand conditions in NKPC estimation (Banerjee et al., 2020). For real GDP growth, a similar methodology has been applied to adjust for the index of industrial production (IIP) and Purchasing Managers’ Index (PMI) composite indicator for India, which are observed every month and are coincident indicators of economic activities (see details of this interpolation in Appendix A2). In addition to these variables, NKPC estimation has been augmented with exchange rate, global commodity prices, financial conditions and global demand conditions to explain the inflation dynamics. On an average, the rupee has depreciated by 4 per cent vis-a-vis USD over the sample period with a very high degree of volatility. Commodity prices have been captured through Indian basket crude oil prices and global non-fuel commodity prices published by the IMF.
To account for financial conditions, Citi financial conditions index (hereafter Citi FCI) for India has been used as a proxy for financial conditions, which has a relatively better forecasting power in predicting real economic activity (Hatzius et al., 2010). The Citi FCI consists of a weighted average of the following variables: corporate spreads, money supply, equity values, mortgage rates, the trade-weighted dollar, and energy prices. A higher value of Citi FCI indicates easy monetary conditions, and a lower value indicates a tight financial condition. To incorporate global demand, US real GDP growth has been considered.
Moreover, we also analyse the importance given to inflation uncertainty and its distributional asymmetry in monetary policy formulation. For this purpose, the weighted average call rate (WACR) has been used as a proxy for the monetary policy rate. WACR is the operating target of the monetary policy, which is stable and symmetrically distributed at around 6.7 per cent.
| Table 1: Summary Statistics | ||||||
| Variables | N | Mean | Std. Dev. | Median | Kurtosis | Skewness |
| CPI-C Headline Inflation | 124 | 6.99 | 3.36 | 5.92 | 2.51 | 0.52 |
| 1-year Ahead Median Household Inflation Expectations | 124 | 11.33 | 2.47 | 11.18 | 1.8 | 0.25 |
| Real GDP Growth Rate | 124 | 6.48 | 1.973 | 6.22 | 6.3 | 1.27 |
| Exchange Rate Growth Rate | 124 | 3.99 | 7.47 | 3.36 | 2.87 | 0.52 |
| Crude Oil Price Growth Rate | 124 | 5.81 | 32.45 | 3.2 | 2.82 | 0.3 |
| Global non-fuel Price Index Growth Rate | 124 | 2.22 | 13.88 | -1.4 | 2.73 | 0.73 |
| US Real GDP Growth Rate | 124 | 2.16 | 0.95 | 2.2 | 12.25 | -2.12 |
| Citi FCI | 124 | -0.39 | 0.38 | -0.46 | 3.11 | 0.67 |
| WACR | 124 | 6.76 | 1.43 | 6.72 | 3.34 | -0.51 |
| Notes: For definitions and sources, please see Appendix Table A1. Source: Authors’ estimates. |
III.3. Methodology
The paper uses quantile regression to estimate the upper and lower tail risks. In a quantile regression, the quantiles of the dependent variable are explained by the set of explanatory variables. In an ordinary least square (OLS) regression, the estimated relationship is the average relationship between dependent and explanatory variables, whereas in the case of quantile regression the estimated relationship is for different specified quantiles. Thus, the benefit of using quantile regression is to estimate the role of explanatory variables in explaining extreme observations (or extreme quantiles) of the dependent variable. Since the study focuses on tail risks of inflation, quantile regression is an appropriate methodology to estimate the tail risks. Putting the quantile regression model more formally, for a dependent variable y explained by realised vector x, the quantile regression for pth quantile is given by,
In the later part of the paper, the conditional distribution has been estimated by minimising the sum of squares of the distance between the estimated quantiles from the quantile regression and quantiles derived from theoretical skewed distribution. The reason behind considering a skewed distribution is to provide space for the possible asymmetric feature of the conditional distribution. The paper considers a skewed normal distribution to estimate the smooth conditional distribution with asymmetric properties. The skewed normal distribution with parameters µ, δ, α, is given by
IV.1. Conditional Distribution and Tail Risks
For analysing the tail risks, the paper considers four domestic factors: one-year ahead households’ inflation expectations based on the RBI survey; real GDP growth; exchange rate vis-à-vis US dollar; and crude oil prices. In addition to this, we have also examined the impact of financial conditions and the role of FIT on tail risks of inflation in the aftermath of its implementation.
All the variables have been incorporated in the form of year-on-year percentage changes in the equations except inflation expectations (which already reflects y-o-y change) and Citi FCI. The FIT framework has been incorporated into the model through a dummy variable which takes the value of 1 after August 2016, and 0 otherwise5. The results of quantile regression are shown in Chart 4. The one-year ahead median household inflation expectations play a crucial role in CPI headline inflation dynamics.
The impact of the inflation expectations is relatively less on lower tail risks and gradually increases for higher quantiles. The domestic real GDP growth has an impact on CPI headline inflation till the third quantile. The exchange rate has a positive effect on tail risks of inflation and a very limited effect at the median level though not significant. However, a time-varying plot of coefficients of exchange rate reveals that it broadly remained significant during 2015-2018 (Appendix Chart A4).
The crude oil price has an impact on the median level and upper tail risks of inflation. Easy financial conditions have a positive influence on tail risks of inflation, and the sensitivity is relatively higher for upper tail risks. The introduction of the FIT dummy identifies the impact of FIT on CPI headline inflation tail risks. A negative coefficient of this dummy supports the success of the FIT framework. Since inflation was on a falling trend even before FIT adoption in India in 2016, the lower as well as the upper tail risks have come down by 3 percentage points.
We estimate the tail risks of inflation based on the estimated parameters of domestic and global determinants of inflation. The tail risks are conditional on historical sample data with 5 per cent tails both in lower and upper tails. The lower tail risk (l(0.05)) is considered as a measure for downside risks to inflation, and the upper tail risk (u0.95) is considered to be upside inflation risks. The estimated tail risks are shown in Chart 5. The upper tail risks of CPI headline inflation fell to a level of around 7 per cent in 2019 from a level of 15 per cent in 2010. Similarly, the lower tail risks also went down to around 2 per cent in 2019 from 8 per cent in 2010. The downward shift in inflation tail risks could be partly attributed to the success of FIT and central bank credibility (Ayres et al., 2014). Importantly, the upper tail risks fell till 2016, and they remained around 7 per cent afterwards, whereas the lower tail risks continuously fell during the sample period. This means that during the FIT period, the downside risks were relatively higher compared to upside risks coming particularly from food reflecting successive years of bumper food grains and horticulture production.
In the scenario analysis, we set preconditions on explanatory variables that influence the inflation dynamics. For simplicity, we assume data points of December 2019 as the preconditions. Next, we consider the standard deviation of the variables for one year before, i.e., January 2019 to December 2019. Then we provide one standard deviation shock to each explanatory variable and examine their impact on the conditional distribution of inflation one by one.
RBI WPS (DEPR): 02/2023: Tail Risks of Inflation in India
https://rbi.org.in/Scripts/PublicationsView.aspx?id=21627
发表回复