India: getting inside the head of the RBI
- In 2016, price stability has become the primary objective of monetary policy of the Reserve Bank of India (RBI). Given the short time span of the newly installed monetary policy framework, it is difficult to forecast the policy rates decisions of the Monetary Policy Committee (MPC). The past year, markets indeed have been struggling to read rates decisions by the MPC
- In this Special, we use a two-equation model to forecast India’s repo and inflation rates. This model contains an adjusted Taylor Rule and Phillips Curve, which are solved simultaneously. The catch of this two-equation modelling approach is that inflationary developments and policy rates decisions are mutually dependent
- In our upper bound forecast, we expect the RBI to start hiking rates from the second quarter of 2018 onwards. This is necessary to counteract rising inflation, caused by a narrowing output gap, higher global inflation and increasing oil prices. The RBI will continue to raise rates until it reaches 7.25% in Q2 of 2019
- In the lower bound forecast, the RBI starts hiking rates in April-June 2019 and follows a more gradual step-by-step approach. Ultimately, the repo rate is kept stable on 7 percent from July-September 2020 onwards. In our lower bound, we are even more hawkish than the repo forecasts from a Bloomberg survey among 46 economists
India has a track record of high inflation. Since the 80s, monthly growth of consumer prices (y-o-y) was 8% on average. High inflation comes with a considerable economic cost, as businesses and households perform poorly when faced with unanticipated large price volatility. Although some studies stress a positive correlation between inflation and economic growth, more recent studies find a negative relationship (e.g. Barro, 2014), which also seems to be non-linear in nature (Eggoh and Khan, 2014). For a fast-growing large emerging economy such as India, it is crucial to keep inflation on a leash and calibrate monetary policy accordingly.
The Reserve Bank of India (RBI) is the body responsible for monetary policy in India. The marriage between the RBI and markets has not always been a happy one. Recently, markets have been struggling to read rates decisions by the Monetary Policy Committee (MPC) of the RBI (see e.g. Bloomberg, 2017a; 2017b), whereas paradoxically the focus of the RBI’s monetary policy has become more transparent ever since it adopted so-called flexible inflation targeting (FIT) in June 2016. The question is if we can get ‘inside the head of the RBI’ and improve our understanding of the MPC’s rates decisions. In order to achieve this goal, in this Special we will discuss a simple model, which provides the groundwork for our repo and inflation forecasts, as it enables us to shed more light on the trajectory of the variables. At the same time, we want to stress that the model outcomes are not our official short-term final forecasts of the repo rate and inflation, which are disclosed in the Monthly Outlook. Rates decisions by the MPC are affected by more factors than captured by the model, so we need a certain flexibility to fine-tune our final forecasts.
A brief history of monetary policy in India
Before we discuss our repo-inflation model in more detail, it is important to elaborate on the changes in monetary policy in India over the course of time. This gives us a better understanding of which variables are important to look at in different phases. We can make a distinction between four phases of monetary policy.
Phase I and II: from fiscal creditor to monetary targeting (MT)
Since the ‘80s, India has shown four distinctive shifts in its monetary policy (see BIS, 2006, IMF, 2017, Figure 1). In the 1980s, India was a closed economy and financial markets were segmented and lacked sophistication. The economy was plagued by volatile and high inflation, which was caused primarily by excessive money supply due to RBI credits to the government. Or as the BIS (ibidem) states it: “monetary policy in India during this period was completely subservient to the fiscal stance of the central government.” In 1985, the RBI adopted monetary targeting with feedback (see Chakravarty Committee, 1985).
The Chakravarty Committee argued that an agreement between the central government and RBI on the extent of monetization of the fiscal deficit was necessary to contain money supply. The money supply growth target rule adopted was not a rigid textbook rule, but also looked at expected output. Meanwhile, fiscal metrics in India kept deteriorating, partly due to the underdeveloped financial system in combination with a ramp up of pre-election spending on several occasions. This resulted in a balance of payment crisis in 1990-91, which raised the sense of urgency to develop and integrate India’s financial markets and improve its banking sector. The coordinating role of the RBI was key in facilitating these changes. From 1992 onwards, lending rates of banks were deregulated, ad hoc T-bills based on monetary financing were phased out to stabilize the fiscal deficit and India adopted a market-determined exchange rate system (Hutchison et al., 2012).
Phase III: multiple-indicator approach and the trilemma
In 1998, the government shifted monetary policy from monetary targeting (MT) to a multiple-indicator approach (MIA). The multi-indicator approach is based on combined information on rates of return in different markets, currency movements, fiscal metrics, capital flows, inflation, refinancing requirements and FX transactions. The move of the RBI to change its policy framework can be understood from two angles (BIS, 2006). First, the RBI and the government made agreements on limiting the possibilities to monetize budget deficits, which reduced the necessity for the RBI to steer on broad money (Q3) as an anchor. Second, the RBI wanted to prepare markets that it was changing its monetary policy from quantity-based signals to price-based signals.
As India’s opened up its financial account, foreign investment inflows increased markedly. Between 2004 and 2007, net investment increased by a factor four, from USD 28bn to USD 108bn. Higher capital inflow leads to higher domestic credit growth and, consequently, economic growth. India’s higher capital account openness also confronted the RBI with a new phenomenon, which is referred to in the economic literature as the ‘impossible trinity’. This means that it is impossible for a central bank to simultaneously maintain: a) a fixed exchange rate, b) independent monetary policy and c) and open capital account. To see why this is the case, suppose India wants to peg the INR to the USD in order to control cost-push inflation. This move would lead to an interest differential with world interest rates. Given the interest parity condition in combination with open capital markets, such a differential would attract foreign capital, which would lead to appreciating pressure on the INR rate. Sterilizing these inflows by buying international reserves and selling INR in order to defend the peg would result in domestic inflationary pressure. Hutchison et al., (2012) show empirically that increasing financial integration of the Indian economy and the corresponding loss of monetary autonomy and exchange rate volatility have influenced inflation and inflation volatility.
Figure 2 shows that surging foreign capital inflows went hand-in-hand with more exchange volatility. To simultaneously safeguard economic, exchange rate and price stability, the RBI continuously had to manoeuver between the three different aspects of the trinity. This explains why the RBI frequently uses foreign reserves to intervene in the FX market and has kept capital controls in place through the Foreign Exchange Management Act (FEMA).
The global financial meltdown after the fall of Lehmann Brothers in 2008 aggravated the policy trilemma faced by the RBI. Crisis-driven expansionary policy was needed to support recovery from the crisis, but in the post-crisis period, the RBI was struggling to keep inflation at bay, given the lack of a clear nominal anchor. Persistent high inflation resulted in negative real interest rates on bank deposits, loss of savings, deteriorating competitiveness and worsening of the trade deficit (IMF, 2017).
Phase IV: flexible inflation targeting
In light of these macro-economic vulnerabilities and taper tantrum in 2013, which resulted in a sliding INR and persistently elevated inflation rates, the RBI decided to review its monetary policy framework once again. In 2015, the government and RBI formally agreed to rollout flexible inflation targeting (FIT) as the new framework and in May 2016, the Reserve Bank of India Act, 1934 was amended, stating that price stability was the primary objective of monetary policy. In addition, a Monetary Policy Committee (MPC) was set up as the body that ultimately decides on the most important policy rates. In August 2016, the nominal inflation target was set at 4% with a bandwidth of +/- 2%.
A new methodology for repo and inflation forecasting
The brief history of India’s monetary policy directly shows the problem to predict future policy moves of the RBI. Inflation targeting is a brand new concept in the Indian economy, so we hardly have any data to work with. For instance, we don’t have any registration of the MPC’s response in case of fast-rising inflation under the FIT regime.
RBI modelling framework
What we do have is information about the framework that the RBI uses. Last year, the RBI has published a core Quarterly Projection Model (QPM) in a working paper covering the FIT framework (RBI, 2016). It is an extensive system of equations which covers India-specific relevant macroeconomic elements, such as the impact of the agricultural sector on food price inflation and an endogenous credibility process for monetary policy. Reproduction of the RBI model could in theory result in very accurate forecasts of the MPC’s rates decision. However, there are three problems which render a full reproduction less useful.
Firstly, the framework is a complex system of equations, assumptions and forecasts on a number of key exogenous variables (i.e. the output gap, agricultural output, foreign prices). Given the complexity of the RBI model, the margin of error of a reproduced model aimed to assess the results that the RBI gets from their model will increase proportionally. This is especially relevant since we don’t have information on the forecast of the RBI on key exogenous variables, such as the output gap, agricultural output and foreign price developments (see also the critique by Rajadhyaksha, 2017).
Secondly, although the RBI has published the framework on its website, it is not clear which members of the MPC are using this framework. This is even more so the question for external members of the RBI. Moreover, although the primary goal of the RBI is pretty straightforward since the implementation of FIT, the MPC composition is far from homogenous. MPC members all have personal preferences in the sense that they attach different weight to output versus inflation stabilization and have different views on their determinants. Hence, using the model outcome of the RBI framework does not guarantee a proper reflection of the rates decision of the MPC, as models are unable to reproduce the discussion between the hawks and doves in the MPC. Third, the forecast horizon of the RBI framework is limited, whereas to make a proper judgment on the medium to long-term trajectory of the Indian economy, inflation and the appropriate rates, we want to use a forecast model that puts the flag further than one or two quarters maximum.
Given the disadvantages of a full reproduction of the RBI FIT framework, we choose a simpler two-equation modeling approach (see the Appendix). As discussed, this model lays the groundwork for our repo and inflation forecasts, but our final repo and inflation forecasts is leading and will be published in the Monthly Outlook prior to each bi-monthly MPC meeting.
The forecast outcome of our model is illustrated in Figure 3. We define an upper band forecast, which shows a tight impulse-response feedback loop between inflation and repo rates. The lower band forecast is based on an econometrically more robust model, but we expect the faster reaction function from the upper bound to produce more accurate forecasts.
The results show that in our upper bound we expect the RBI to start hiking rates in the second quarter of 2018 and we expect that it will continue to follow an upward trajectory until rates have reached 7.25% in Q2 of 2019. Inflation will only slightly breach the 6% upper bandwidth. The RBI will have to start hiking again in 2021 to keep inflation below the 6% bandwidth. Our lower bound shows that the RBI starts hiking rates in Q1 of 2019 and follows a more step-by-step upward trajectory. The repo rate will be kept on 7% from 2020Q3 onwards.
Interestingly, results from a Bloomberg survey in November asking 46 economists about their forecasts for India’s economy shows that the median expectation of the repo rate will remain rock solid at 6% until Q2 of 2019. Even in our lower bound scenario, we expect that rates will be hiked in 2019Q2 and, most likely, much sooner.
Inflationary pressure on the rise
The inflationary pickup that the model translates into repo rate hikes is especially due to rising global inflation, narrowing of the output gap and a further increase in commodity prices, such as oil (Figure 4). Although we expect the INR rate to be quite stable (and even appreciating slightly against the USD), a faster than anticipated tightening cycle by the Fed or the ECB will also lead to additional inflationary pressure in India.
Despite the introduction of price stability as its primary goal in 2016, markets have experienced difficulties reading the policy changes of Reserve Bank of India (RBI). The question is whether we can get ‘inside the head of the RBI’ and improve our understanding of the MPC’s rates decisions. As flexible targeting (FIT) is a brand new concept in India, we don’t have much empirical data to work with. We do have the analytical framework for FIT, but it is not useful to reproduce this complicated system of equations without substantially increasing the margin of forecasting error.
Therefore, in this Special we have adopted a simple inflation-repo rates forecasting model, which is based on an adapted Taylor rule and a Phillips curve. We use an upper and lower bound forecast of the repo rate. The upper bound forecast is based on a tighter impulse-response feedback loop between inflation and repo rates than the lower bound forecast. In our opinion the upper bound model produces more accurate forecasts, whereas the lower bound is based on a more robust econometric model.
The results from our model show that in our upper bound forecasts, we expect the RBI to start hiking rates in the second quarter of 2018. This is necessary to counteract rising inflation, caused by a narrowing output gap, higher global inflation and increasing commodity prices. The RBI will continue to raise rates until it reaches 7.25% in Q2 of 2019. In our forecast, inflation will only slightly breach the 6% upper bandwidth. The RBI will have to start hiking again in 2021 to keep inflation below the 6% bandwidth. In lower bound forecast path, the RBI starts hiking rates in 2019Q2 and follows a more gradual step-by-step approach. Ultimately the repo rate is kept stable on 7% from 2020Q3 onwards. Even in our lower bound, we are more hawkish in our expectations than the median of 46 economists’ forecasts in a Bloomberg survey dated from November 2017.
We underline that our model is far from flawless and it will require refinement over time. Nevertheless, it provides the basic groundwork for our repo and inflation forecasts and enables us to shed more light on the trajectory of these variables. It is important to stress, however, that our final forecast on repo and inflation rates is leading and will be published in the Monthly Outlook prior to each bi-monthly MPC meeting.
Barro, R.J. (2013). Inflation and Economic Growth. Annals of Economics & Finance, 14(1).
Benes, M.J., K. Clinton, A. George, P. Gupta, J. John, O. Kamenik, D. Laxton, P. Mitra, G.V. Nadhanael, R. Portillo, H. Wang and F. Zhang (2017). Quarterly projection model for India: key elements and properties. International Monetary Fund.
Bhattacharya, K. (2006). Monetary policy approaches in India. Bank for International Settlements (BIS).
Eggoh, J. C., & Khan, M. (2014). On the nonlinear relationship between inflation and economic growth. Research in Economics, 68(2), 133-143.
Gordon, R.J., & J.H. Stock (1998). Foundations of the Goldilocks economy: supply shocks and the time-varying NAIRU. Brookings papers on economic activity, 1998(2), 297-346.
Government of India (2015). Agreement on monetary policy framework between the government of India and the Reserve Bank of India.
Government of India (2016). Amendments to The Reserve Bank of India Act, 1934. Chapter XII, Miscellaneous, Part I, The Finance Act 2016, pp. 82-87.
Hutchison, M., R. Sengupta and N. Singh (2012). India’s trilemma: financial liberalisation, exchange rates and monetary policy. The World Economy, 35(1), 3-18.
Marey, P. (2017). What’s wrong with Taylor rules? Rabobank.
Rajadhyaksha, N. (2017). Adding heft to inflation targeting. LiveMint, May 2017.
Reserve Bank of India (2016). Inflation-forecast targeting for India: an outline of the analytical framework. RBI Working Paper Series No. 07.
Rudebusch, G.D. (2002). Term structure evidence on interest rate smoothing and monetary policy inertia. Journal of monetary economics, 49(6), 1161-1187.
Vinayagathasan, T. (2013). Inflation and economic growth: A dynamic panel threshold analysis for Asian economies. Journal of Asian Economics, 26, 31-41.
Virmani, V. (2004). Operationalising Taylor-type rules for the Indian economy. Technical report, Indian Institute of Management.
Appendix 1: Technical aspects of our repo and inflation model
In this appendix we elaborate on the technical aspects of our two-equation model. It is based on two well-known principles in the economic literature: the Taylor rule and the Phillips curve. Below we discuss each element separately.
In order to estimate interest rates, often a simple Taylor rule is used (Taylor, 1993):
where i* is the interest rate or, in our case, the repo rate, r* is the equilibrium real interest rate, p is the inflation rate (y-o-y percentage change of consumer price index or CPI), p* is the target inflation rate, y is the output gap and t is a time-specific index. The output gap is defined as:
where Y is real gross domestic product (GDP) and Y* is potential GDP. We use two indicators for potential GDP. Our first measure is a HP-filtered trend of GDP and the second is potential output measured by the IMF.
A problem with the original Taylor rule is that it does not take into account inertia in the policy rate setting by the central bank. If we apply the original rule, the estimated fit of the policy rate would show high volatility, as both the output gap and inflation fluctuate heavily on a quarterly basis. To solve this problem, we introduce policy rate smoothening:
where the central bank gradually adjusts its repo rate i to the desired rate i*. Ultimately we get the following equation (see Virmani, 2004):
Table 1 shows our estimation results. Column (1) shows the OLS estimation results using quarterly data for the entire sample (2000Q1-2017Q3). All variables show the expected sign, but only the output gap is significant, which means that during the MIA period the RBI policy decisions were mainly driven by growth consideration rather than price stability. If we use the output gap of the IMF, we get the same results (see column 2 in Table 1). Interestingly, if we narrow the sample to start from 2014, the period where the RBI was actively anticipating on implementing FIT, a completely different picture emerges (column 3 in Table 1). The output gap does not have a significant effect anymore, but inflation does have a significant and positive effect. The variety of our estimation results immediately illustrate that it is important to take into account the various stages of monetary policy that India went through over the course of time.
There are two problems that make forecasts using equation (4) less reliable. First, the policy response of the RBI in equation (4) is linear, regardless of the inflationary situation in the economy. This is an unrealistic assumption, as the RBI would act increasingly forceful when inflation is further away from the policy target. Therefore, we adjust equation (4) as follows:
The estimation of equation (5) shows that the policy response of the RBI is stronger in case of breaches of the 2% and 6% bandwidth (column (4) of Table 1). As a substitute for the constant term, we have included the level of potential growth, which functions more or less as a trend variable. The inflation coefficients are no longer statistically significant, but this is especially due to the limited estimation sample.
The second problem with a single equation repo model is that we require a static forecast of inflation, whereas inflation itself is also affected by changes of the repo rate. Therefore, we need a second inflation equation, which we need solve together with equation (5) in order to obtain simultaneous inflation and repo forecasts.
Starting point for our inflation equation is the adjusted Phillips curve model by Gordon (1997), who uses three basic determinants of the inflation rate: inertia, demand and supply:
where p indicates inflation, D is a measure of excess demand (in our case the output gap: y) and Z denotes supply-side shocks, such as oil price shocks or exchange rate movement. Ultimately, we arrive the following equation for India:
where c is a constant term, et is mutation of the INR exchange rate vis-à-vis the USD, pw is mutation of the global price level and i is the repo rate. All mutations are expressed in year-on-year percentages.
The estimation results of equation (7) are shows in Table 2. We again use OLS on a quarterly basis, but we start with a longer estimation sample (2001Q2-2017Q3). Column (1) shows the full model estimate. All variables show the expected sign. The smoothing parameter shows that 85% of inflation in the previous quarter feeds into the current quarter. The output gap has a positive (albeit insignificant) effect, indicating that if domestic demand picks up, inflationary pressure also rises. The exchange rate (INR per USD) shows a positive effect. An increase of variable means that you get more INR per USD, which implies that the INR has depreciated. A depreciation leads to higher import inflation, which pushes up domestic inflation. Higher global prices lead to higher domestic inflation as well, but it is possible that this effect is partly captured by the exchange rate variable. This might explain why do not find a significant effect of this variable. Finally, the repo rate is, as expected, negative, but again we do not find a significant impact. Our results improve when we use the IMF output gap in column (2) of Table 2, instead of the HP filtered output gap. The magnitude of the coefficients for the output gap and the repo rate are slightly higher and statistically significant, whereas the coefficients of all other variables remain stable. The only variable that does not show a significant effect is global inflation.
When we limit our estimation sample to 2008Q1-2017Q3, the output gap shows a counterintuitive sign, or in case of the IMF output gap an insignificant sign (column (3) in Table 3). If we substitute the elasticity of the output gap in column (4) by the elasticity found in column (2), we end up with practically the same results as the ones found with the IMF output gap in column (3). However, in column (4), global prices also have a significant impact on the repo rate, though at a lower confidence level of 90%.
A disadvantage of all estimated models is that the speed with which repo rate mutations feed into inflation is exceptionally gradual. Although a certain time lag is not unrealistic, we expect the RBI to respond much more aggressively in case the bandwidth targets, especially the upper bound, is threatened to be breached. Therefore, in column (5) we calibrate the elasticity of the repo rate by using a upward scaling factor on the output elasticity, whereas at the same time we use a downward scaling factor on the overall level of inflation. Although this estimation is perhaps less robust from a statistic and econometric point of perspective, it has a more realistic angle in our opinion. Ultimately, we use estimation in column (2) as our most pure lower-bound estimation of the repo rate and we use the model in column (5) as our upper-bound estimation, which probably will produce more robust forecasts. The inflation fit of both model is shown in Figure A.1.
One of the major flaws of our repo-inflation model is that it does not contain forward looking and money supply elements. We have experimented with both inflation expectations as well as the money supply, but this did not generate any fruitful results. Moreover, it is well-known that the RBI does not only look at CPI, but also at core inflation (i.e. inflation minus food and fuel prices), in order to take into account so-called ‘stickiness’ of inflation. The problem with core inflation is that we only have limited time series on food and fuel prices. Lastly, we do not separately examine food price developments, whereas these make up for 45% of total price fluctuations. Hence, factors such as the monsoon are very important for inflation, but the amount of precipitation is impossible to predict. As said, our current model provides the rough groundwork for our repo and inflation expectations, but further refinement of the model are needed to make sure the groundwork will evolve into a firm house in the future.