I. Leading Indicators for Inflation in Russia1
The focus of monetary policy in Russia is shifting towards inflation targeting. This note develops two distinct but closely related analytical tools to inform monetary policy: trimmed mean core inflation and a leading indicators model (LIM) for inflation forecasting. The trimmed mean core inflation measure tracks trend inflation in Russia better and is less volatile than the Russian Federal Service of State Statistics (Rosstat) core inflation measure. This core inflation measure indicates that the recent surge in headline inflation is not entirely attributable to food price shocks as broad inflationary pressures are also evident. The LIM identifies a group of leading indicators that best fit Russia’s headline and core inflation dynamics during 2003–11. The model suggests that headline inflation is strongly associated with past developments of broad money and food prices. These findings suggest that inflation at end-2011 will remain well outside the CBR’s targeted range of 6–7 percent.
1. The focus of monetary policy in Russia is shifting toward low and stable inflation. Like many other central banks, the Central Bank of Russia (CBR) has been charged with several objectives—ensuring a stable exchange rate, low inflation and, in practice, supporting high growth.2 These objectives can be difficult to achieve simultaneously with the exclusive use of monetary policy instruments. When facing policy conflicts over the past few years, the inflation target was often compromised to achieve other objectives. However, there has been a notable shift in the focus of monetary policy recently toward greater exchange rate flexibility and stronger commitment to inflation targeting. Relevant for the transition toward full-fledged inflation targeting, this chapter conducts two distinctive but closely related analyses of core inflation and inflation forecasting in Russia.
2. Specifically, this chapter seeks a set of leading indicators for inflation. As monetary policy affects inflation with long and variable lags, central banks should take a view on future evolution of inflation to ensure that the intended effect of policy decisions materializes at the right time. In this respect, finding a stable empirical relation between current data and future (trend) inflation would help inform monetary policy decisions. The remainder of the note is organized as follows. Section B proposes a new measure of core inflation, which estimates the trend component of monthly inflation using real-time data. Section C presents short-term inflation forecasts for Russia derived from a LIM. Section D concludes with some policy implications.
B. Core Inflation
3. Core inflation is meant to be a good indicator for trend inflation and a viable target for monetary policy. A measure of core inflation usually smoothes out temporary price fluctuations to uncover the trend component of inflation. By allowing policymakers to see through temporary or potentially misleading fluctuations, core inflation measures can guide the direction of monetary policy. Further, as temporary price fluctuations are often caused by non-monetary forces, core inflation is generally considered to be more controllable by monetary authorities than headline inflation. Given these advantages, core inflation is closely monitored and often used as an implicit monetary policy target by central banks.
4. A core inflation measure can also improve policy effectiveness by providing a useful tool for transparent public communication. Core inflation could help clarify why policymakers are or are not reacting to fluctuations in headline inflation rates. Public communication through a core inflation measure would also direct public attention to trend inflation, bringing public focus in line with the focus of the monetary authorities. This would be important for Russia: given the highly persistent inflation in the past, successful inflation targeting would require anchoring inflation expectations. To the extent that the focus on core inflation reduces the pass through of temporary shocks to inflation expectations, the variability of inflation would be further reduced.
5. The main objective of this section is to develop an estimate for trend inflation with “real-time” data. As transitory forces can only be known with the benefit of hindsight, the true trend inflation cannot be recovered with certainty until after the fact. A variety of methodologies are used for the computation of core inflation, as an estimate for trend inflation, reflecting country-specific circumstances. There are three types of core inflation measures, depending on how volatile price movements are smoothed out.
- Exclusion method: the core CPI of this kind excludes a predetermined list of CPI components—typically, volatile (and seasonal) food and energy prices. Some central banks exclude “administered” services prices, reflecting country-specific circumstances.
- Trimmed mean method: the core CPI of this kind excludes a fraction of the most extreme price movements in both tails of the “cross-section” price distribution. This method shares the same idea with the exclusion method in the sense that it leaves out more volatile short-term price movements. However, unlike the former, the trimming is purely statistical, and the CPI components that are trimmed vary over time.
- Moving average method: the core CPI of this kind is calculated as a moving average of past monthly headline inflation rates.
6. The estimation method and the use of core inflation measures differ across countries. The Russian Federal Service of State Statistics (Rosstat) compiles a core inflation measure using an exclusion method and publishes it on a monthly basis along with headline inflation. Currently, the exclusion method is more widely used than the other two methods, as it is more transparent and easier to communicate to the public. However, the trimmed mean method is also widely used for analytical purposes and as a robustness check of core inflation measures (Box 1).
Box 1.Core Consumer Price Index in Selected Countries
Rosstat calculates core CPI by excluding a predetermined list of goods and services—fruit and vegetable, fuel, and administered service prices—from the headline index. The share of excluded items in the 2005 CPI basket was 21 percent. The core inflation rate is published on a monthly basis along the headline inflation and its breakdown for food, non-food goods, and services. The CBR’s end-year inflation target is set in terms of headline inflation.
The Central Bank of Brazil estimates core inflation as a trimmed mean, leaving out 20 percent of weights from both tails, which scores the best fitting with 13-month centered moving average of monthly inflation. Core inflation is published in the quarterly Inflation Report.
The Bank of Canada (BOC) calculates core inflation by excluding food and energy prices and the effects of changes in indirect taxes from the headline index. While formal inflation targets are expressed in terms of the headline CPI, the BOC explicitly focuses on the core measure in seeking to implement the targets. The trimmed mean method is also used, though not formally adopted, for robustness check and analytical purposes.
The Central Bank of Turkey publishes two core inflation measures in its quarterly Inflation Report. Core inflation measures are calculated using the exclusion method and called H or I core, depending on the items excluded. H core inflation excludes unprocessed food products, energy, alcoholic beverages, tobacco, and gold; I core inflation excludes broader food items than H core inflation. Other methods, including the trimmed mean method, are used for robustness check.
The U.S. Federal Reserve compiles core personal consumption expenditure (PCE) price index by excluding volatile and seasonal food and energy prices from the PCE price index, which is a U.S.-wide indicator of the average increase in prices for all domestic personal consumption. Some Federal Reserve Banks uses the trimmed mean method (Dallas) or the moving average method (New York) for the calculation of core inflation.
Other central banks that calculate core inflation by excluding food and energy prices include the European Central Bank, the Reserve Bank of India, and the Bank of Korea.
7. The trimmed mean method of a core inflation measure has several important advantages. Given a secular change in the relative price of volatile and seasonal items (e.g., food), the trimmed mean method ensures that estimated core inflation moves closely with trend headline inflation. This is different from the exclusion method, where there could be persistent gaps between trend inflation and core inflation.3 Further, the trimmed mean method is flexible in handling the skewness of the cross-section price distribution. Given that a typical cross-section distribution of component price changes is skewed to the right—meaning headline inflation tends to be more influenced by extreme positive price changes, the asymmetric trimming of extreme values would ensure smoother core inflation than the moving average method, which implicitly assumes symmetry of extreme price movements.
|Exclusion method||Easy to calculate, understand, and communicate (e.g., core = headline net of food prices).||May fail to exclude some highly volatile components, while throwing out some useful information.|
Would have a persistent gap from trend inflation when there is a secular change in the relative prices of the excluded items.
|Trimmed mean method||Maximum use of information and less room for discretion.|
Trimming criteria will be derived as an optimal solution.
|More challenging to calculate, understand, and communicate.|
|Moving average method||Easy to calculate. Transparent.||Backward looking or only available well after the fact.|
Could be more volatile than trend inflation if the cross-section price distribution is skewed.
Data and methodology4
8. Based on the trimmed mean method, this note estimates core inflation in Russia. Disaggregated CPI components and the corresponding weights for the period from January 2005 to May 2011 are used for the estimation. Seasonally-unadjusted series are used for the baseline cases, while seasonally-adjusted series are also examined for robustness tests.5 The seasonal adjustments are made using X12-ARIMA.
9. Specifically, two trimming methods are used in this note which turn out to generate similar results.
- Fixed-weight approach drops extreme values of a certain percentage of weights from each tail of the cross-section price distribution of 46 CPI components in each month. As the typical price distribution is skewed, the cutoff weights are not constrained to be symmetric.
- Standard deviation approach drops the extreme values that are a certain standard deviation away from the average in each month. This trimming does not need to be symmetric, either.
As these trimming methods leave out most volatile price movements in each month, the list of excluded CPI components varies each month. Further, unlike the fixed-weight approach, where the weights of excluded components are fixed throughout the sample period, the excluded weights under the standard deviation approach vary each month. This flexibility makes the standard deviation approach trim less information than the other approach to generate a smooth measure of core inflation.
|Components excluded each month||Weights of excluded components|
|Standard deviation approach||Varying||Varying|
10. The trimming points are chosen to minimize the root mean square distance between the trimmed mean and trend inflation. The proxy for the trend inflation is a centered 24-month moving average of seasonally unadjusted monthly inflation rates.6 As seen in Figures 1–2, the centered moving average is quite stable, and averaging a longer time span makes little difference in the optimal trimming points.
Figure 1.CPI Inflation, Fixed-Weight Approach Figure 2.CPI Inflation, Fixed-Std Dev Approach
11. The fixed-weight approach trims out 23 percent of weights from the right tail and 41 percent from the left tail. Consistent with our prior, the “most-often-excluded” components for the sample period include food, energy, and administered prices such as fruits and vegetables, eggs, sugar, gasoline, and passenger transport. However, communication devices and services are also frequently trimmed from the left tail, possibly reflecting the effect of fast technological progress and more intense competition.
12. Core inflation based on the fixed-weight approach suggests that trend inflation has been picking up considerably since August 2010 (Figure 1). Following a significant decline from 14.5 percent in March 2009 to 5.3 percent in July-August 2010, the 12-month core inflation rate increased gradually to 8.2 percent in May 2011. This increase in core inflation is less striking than the acceleration of headline inflation from 5.4 percent in July 2010 to 9.6 percent in May 2011. However, the evident upward trend in the core measure indicates that the acceleration of headline inflation was not entirely attributable to supply shocks to food prices.
13. The standard deviation approach generates similar results (Figure 2). The 12-month core inflation increased from 5.3 percent in July 2010 to 7.8 percent in May 2011, suggesting a rising trend inflation. The optimal trimming under this approach drops price movements that are 1.6 standard deviation above the sample mean (from the right tail) and 1.0 standard deviation below the mean (from the left tail). For the sample period, on average, 4.1 percent of weights are trimmed from the left and 4.5 percent from the right (total 8.6 percent). The top 10 “most-often-excluded” items include fruits and vegetables, eggs, sugar, other food, gasoline, communication devices, cheese, health rehabilitation services, other services, pasta and cereals—mostly food items with a few service and nonfood items.
|Ranking||Item||Weight (‘11)||Ranking||Item||Weight (‘11)|
|1||Fruits and vegetables||4.18||6||Communication devices||0.50|
|3||Sugar||0.70||8||Pasta and cereals||0.96|
|4||Other foods||2.38||9||Health services||0.49|
14. Core inflation measures based on the trimmed mean methods are consistently less volatile than Rosstat’s measure throughout the sample period (Figures 3–4). Despite the large gap in the amount of information used for the estimation—the fixed-weight approach trims 64 percent of total weights while the standard deviation approach trims only 8.6 percent on average, the two trimming methods generate remarkably similar results. However, as can be seen from the peak-to-trough variations of year-on-year rates and the short-term volatility of monthly rates, the Rosstat’s core inflation is more volatile than the trimmed mean inflation and sometimes more than the headline inflation. This high volatility makes Rosstat’s core inflation less attractive as a leading (or real-time) indicator of trend inflation.
Figure 3.Core Inflation Rates, 12-month rate Figure 4.Core Inflation Rates, month-on-month rate
15. Core inflation can inform monetary policy by revealing the evolution of trend inflation. The proposed core inflation measures indicate that the recent surge in headline inflation is not entirely attributable to food price shocks, as broader inflationary pressures are also evident. This suggests that monetary tightening would be needed to bring headline inflation under control. As a next step, an inflation forecasting model is presented to provide a better understanding of the relationship between current economic variables and future inflation. The model provides a more focused view of future inflation dynamics, and policy implications for inflation targeting.
C. Inflation Forecasting: Leading Indicators Model
16. A more explicit econometric analysis for short-term inflation forecasting is a useful tool to inform and guide monetary policy. This section identifies a group of leading indicators for monthly inflation rates between 2003 and 2011. Estimated empirical relations between the leading indicators and inflation are then used to project inflation 6–12 months ahead.
17. LIMs are widely used for inflation forecasting. LIMs rely on empirical correlations between selected economic variables and actual inflation, and do not impose explicit causal relationships between them. This flexibility improves the forecasting accuracy in some cases, but at the cost of not establishing structural relationships and thus making policy implications less tractable. For this reason, LIMs are often used as a complement to a fully-fledged structural model estimation, which explicitly addresses issues relating to the monetary policy transmission mechanism.
18. However, it is well accepted that LIMs are particularly useful in detecting turning points in inflation. LIMs translate turning points of leading indicators into those of future inflation. This study finds that LIMs accurately capture the timing of sustained upward and downward movements in the headline inflation rate in February 2007, July 2008, and June 2010.
19. An autoregressive distributed lag model (ADL) is estimated for the period from July 2003 to April 2011. The general-to-specific algorithm in Ox Metrics is used to narrow down the set of possible leading indicators from a larger dataset. Then, various metrics, including significance tests, statistics measuring the ex-post forecast quality, and consistency with economic theory, are considered to choose the benchmark model. The sample period is constrained by a structural break and availability of key variables. Separate models for headline and core monthly inflation are estimated.7 The results of the two regressions are presented in Table 1 and Table 2 below.
|lagged 3 months||0.06||0.05||1.21|
|lagged 6 months||0.17||0.05||3.82|
|lagged 1 months||-0.03||0.01||-4.41|
|lagged 2 months||-0.03||0.01||-3.96|
|lagged 4 months||-0.03||0.01||-2.99|
|lagged 6 months||-0.03||0.01||-2.93|
|lagged 7 months||0.03||0.01||3.67|
|lagged 12 months||0.02||0.01||3.28|
|lagged 0 months||0.32||0.02||14.9|
|lagged 1 month||0.25||0.04||6.78|
|lagged 5 months||0.06||0.04||1.48|
|lagged 9 months||0.06||0.04||1.67|
|lagged 13 months||0.09||0.04||2.50|
|lagged 1 month||-0.07||0.01||-5.28|
|lagged 3 months||-0.04||0.02||-2.73|
|lagged 5 months||-0.04||0.02||-2.98|
|lagged 7 months||-0.02||0.02||-1.03|
|lagged 9 months||-0.03||0.02||-2.02|
|lagged 11 months||-0.04||0.02||-2.71|
|lagged 5 months||0.05||0.01||3.4|
|lagged 9 months||0.02||0.01||1.79|
|lagged 12 months||0.03||0.02||1.98|
|lagged 13 months||0.01||0.01||0.66|
20. The following variables are considered as potential leading indicators, in line with other empirical studies:
- Interest rates: Interbank 3-month offer rate, 5-year generic bond yield, deposit rate;
- Asset prices: housing prices, equity price index;
- Real activity: real GDP, industrial production index, unemployment rate;
- Monetary aggregates: Monetary base, M2, private sector credit;
- External variables: US Federal Fund rate, world commodity prices (food and beverages), partner country inflation, crude oil prices, gold prices;
- Exchange rates: RUB/$ exchange rate, NEER;
- Others: General government revenue, average wages, business confidence index.
21. Lagged headline inflation, broad money growth, nominal effective exchange rate and food price inflation are leading indicators of headline inflation (Table 1). Broad money growth (M2) is positively correlated with inflation with 7 to 12-month lags, which is broadly in line with other empirical studies on Russia. Exchange rate pass-through also affects inflation significantly with a lag of 1-6 months. Food inflation is estimated to have an immediate impact on headline inflation, with the estimated coefficient close to its weight in the CPI basket.
22. The same set of leading indicators is found for core inflation (Table 2). The lags between M2 and core inflation are similar to those found between M2 and headline inflation. However, the exchange rate pass-through to core inflation is estimated to be larger and more persistent than that to headline inflation. Food inflation also has a persistent effect on core inflation, possibly reflecting its impact on inflation expectation and second-round effect on core inflation.
23. The 12-month ex-post forecasts correctly predict the turning point of both headline and core inflation in August 2010.8 While the 12-month forecast for core inflation underestimated the core inflation rates throughout the forecasting horizon, the headline inflation forecasts predict the actual inflation very closely (Figure 5). Forecasting accuracy for core inflation improves in a shorter-forecasting horizon. The LIMs predict the rising headline and core inflation in the second half of 2010 with a remarkable accuracy (Figure 6).
Figure 5.Twelve-Month Ex-Post Forecasts Figure 6.Six-Month Ex-Post Forecasts
24. The LIMs project that headline inflation will start to decelerate from August 2011 (Figure 7). The projected turning point reflects the assumed M2 growth, as well as the base effect of high inflation in the second half of 2010. Specifically, the out-of sample forecasts are based on the following assumptions for 2011: (i) food price inflation at 7.8 percent with favorable weather conditions, (ii) NEER appreciation at 3.0 percent, and (iii) M2 growth at 25 percent.9
Figure 7.Inflation forecasts for 2011
25. The model predicts headline inflation at 8.0 percent and core inflation at 7.9 percent at end-2011 (Figure 7), with a relatively wider margin of error for core inflation forecasts. The model suggests that food price inflation is the key risk to the inflation outlook (Figure 8). When Russia’s food prices are assumed to increase at the same pace as the world food prices (21.8 percent in the May 2011 WEO Global Assumptions), both headline and core inflation are projected to keep rising in 2011 to double-digits.10
Figure 8.Inflation forecasts for 2011 with WEO Global Assumptions for food prices
D. Policy Implications
26. Both core inflation and LIMs suggest that the recent surge in inflation is not attributable only to food price increases since the summer of 2010. The proposed estimates of core inflation indicate that the rising headline inflation in recent months has been triggered mainly by food inflation, but broader inflationary pressures are also evident. LIMs also suggest that the surge in inflation is strongly associated with the past developments of monetary aggregates.
27. These findings suggest that inflation at end-2011 will remain well outside the CBR’s targeted range of 6–7 percent. While stable food price is key to lowering inflation, some moderation of M2 growth—through limited intervention in the foreign exchange market and higher policy rates—will be needed to bring inflation under control. However, given the 7–12 month transmission lags, the effect of policy tightening would likely be felt only in 2012. In addition, it should also be noted that there is limited scope to use LIMs to analyze monetary policy in more detail, particularly, due to the lack of explicit considerations on the monetary policy transmission mechanism. In this respect, further studies with a structural model are warranted to investigate causal relationships between inflation and other economic and policy variables.
The estimation of core inflation uses disaggregated CPI series and their weights for 46 items for the period from January 2005 to May 2011 (15 foods, 19 nonfood goods, and 12 services items). The weights in the CPI basket are revised each year. The average weight for an item in the CPI basket is 2.2 percent, with the maximum of 10.7–9.6 percent for meat and the minimum of 0.1–0.2 percent for hospitality service.
Statistical Properties: Cross-Section
Data allows us to examine the cross-section distribution of 46 CPI components in each month for the sample period. When
- Kurtosis: Higher kurtosis implies fatter tail, meaning greater influence of extreme values. This establishes the usefulness of the trimmed mean.
- ➢ Over the sample period, the average kurtosis is 1,887 for seasonally unadjusted (cross-section) distributions and 1,413 for seasonally adjusted distributions. This is very large, suggesting greater influence of extreme values on Russia’s inflation dynamics. For the U.S. during 1977–2004, the average kurtosis of monthly inflation was 40.6.
- Skewness: Positive (negative) skewness implies longer right (left) tail. The presence of skewness is not essential to the statistical case for trimming, which is based on the presence of excess kurtosis. However, a finding of skewness suggests that we should not constrain our trim to be symmetric.
- ➢ Over the sample period, the average skewness is 39 for both seasonally unadjusted and seasonally adjusted distributions. (For the U.S. during 1977–2004, the average skewness was 0.36.) This implies that on average, monthly inflation is affected more by positive extreme values. However, it is fickle with positive skewness in 61 percent of the sample months and negative skewness in the rest. Seasonal adjustment makes little difference in this pattern. The large variation of skewness reflects volatile food items and administered service prices as well as the small number of components—in other country cases, much more detailed breakdown of CPI is used, which usually improves the stability of price distributions.
Statistical Properties: Time-Series
Standard deviations of the time series of each component reveal which items in the CPI basket drive the swings of monthly headline inflation. In general, food inflation is most volatile, as we expected, followed by service inflation.
This pattern survives seasonal adjustments of each series, implying seasonal effects are not the main forces driving higher volatility of these prices.
More accurate description of the trimming methods proposed in Section B is as follows:
- Fixed weight approach: Drop α percent of the weights from the left tail of each month’s distribution and β percent of the weight from the right tail.
- When the N components of CPI are ordered such that π1t ≤ π2t ≤ … ≤ πNt with the corresponding weights (ωit), let
. Then, the trimmed inflation is
- ➢ Standard deviation approach: Drop the components that are X standard deviation below the average inflation in each month and p standard deviation above the average inflation.
, the trimmed inflation is , .
Then the optimal trimming chooses the parameters or (α, β) or (λ, ρ) to minimize the distance from the trend inflation, i.e.
BokilM. and A.Schimmelpfennig2005 “Three Attempts at Inflation Forecasting in Pakistan” IMF Working Paper No. 05/105 (Washington: International Monetary Fund).
BanerjeeA. M.Marcellino and I.Masten2005 “Leading Indicators for Euro-area Inflation and GDP Growth” Oxford Bulletin of Economics and Statistics 67 pp. 785–813.
DolmasJ.2005 “Trimmed Mean PCE Inflation” Research Department Working Paper No. 0506 (Dallas: Federal Reserve Bank).
HoffmaisterA.W.2001 “Inflation Targeting in Korea: An Empirical Exploration” IMF Staff Papers Vol. 48 No.2 pp. 317–343.
HoganS. M.Johnson and T.Lafleche2001 “Core Inflation” Technical Report No. 89 (Bank of Canada).
PhilipA. K.1986 “Leading Indicators of Inflation in Market Economies” International Journal of Forecasting Vol. 2 No. 4 pp. 403–412.