Abstract
This Selected Issues paper provides background information and analysis on recent developments and critical issues for the Colombian economy. The study discusses the unemployment and stresses in the financial system and also focuses on fiscal issues. The following statistical data are presented in detail: national accounts at current prices and at constant prices, savings and investment, value of agricultural crops, mining production, structure of regular gasoline prices, indicators of construction activity, minimum wages, producer price index, interest rates, and so on.
V. Core Inflation in Colombia49
A. Introduction
75. During the 1990s, several developed economies have shifted their approach to monetary policy from strategies that focused on intermediate monetary targets (like specific monetary aggregates or exchange rates) to inflation targeting.50 Under this approach, the monetary authority focuses its policies on achieving a pre-selected inflation target. Given that an inflation targeting policy makes inflation—rather than economic activity, employment, or any other criterion—the key objective of central bank policy, it helps countries that do not have a good record of inflation control to strengthen their credibility in this area.51 While having succeeded in reducing inflation during the 1990s, in 2000 the Colombian authorities expressed their commitment to subordinate the intermediate monetary targets to the inflation goal.
76. In the context of inflation targeting, the choice of the price index that becomes the relevant target for monetary policy is a fundamental one, as central banks aim to affect the part of price variations that is linked to monetary phenomena. Price indices, such as the consumer price index (CPI) may not be a suitable indicator for the purpose of inflation targeting, since it is affected by both demand and supply shocks. Thus, sector specific shocks could temporarily introduce price movements that would make it difficult to isolate the behavior of prices in the long run. Core inflation indicators can play a fundamental role to eliminate short-term price movements.52 This note aims to analyze the behavior of core inflation indicators in Colombia during the 1990s.
77. Core inflation indicators (CIIs) for Colombia were previously analyzed in Jaramillo et al. (1999). In that work, the authors studied the statistical properties of different core inflation indices for the period from 1988 to 1998. These properties correspond to important characteristics that an index of monetary inflation should have. Specifically, previous works focused on the following issues: (i) volatility; (ii) bias; (iii) forecasting power; and (iv) capacity of monetary aggregates to explain its behavior. This note will: (a) follow the methodology used by Jaramillo et al., and extend the sample of analysis of several CIIs, and (b) add to the analysis three new CIIs, the 12-month moving average inflation rate, the 24-month moving average inflation rate, and the 12-month moving average of the asymmetrically trimmed indicator.53 We will show that the behavior of the latter indicator could make it an appropriate variable for monetary authorities to monitor.
B. Core Inflation Indicators
78. There are two main kinds of CIIs. A first sort of indices is called “limited-influence indicators.” The basic feature of these indicators is that the existing index (typically the CPI) is reweighted by placing zero weights on some components. A second sort of indices takes into account all the components of a relevant price index and uses a statistical filter to obtain the trend of the inflation rate.54 We analyze the behavior of seven CIIs that can be classified in three groups:
The first group is given by limited-influence indicators that eliminate from the index a regular bundle of components. These components do not change from period to period. We focus on two indicators of this kind. One is the CPI without food products. The justification for this kind of indices is that price variations in food products tend to be volatile and strongly influenced by supply shocks. A drawback with these sorts of indices is that the selection criteria of the goods to be eliminated may not be appropriate in certain cases.55 The second index is called the fixed trimmed weights 20 indicator (Inflación Núcleo 20). This index is built by dropping a fixed set of components from the CPI. A period is selected and the most volatile components in the index are identified.56 These are excluded until their cumulated weight in the distribution sums to 20 percent. The computation of this index is very simple. The drawback is that it forces the analyst to review periodically the bundle of goods to be eliminated from the index.
The second group includes limited-influence indicators that eliminate different bundles of goods in different periods. We use two indices from this class. One is the trimmed mean 10. This indicator is constructed by averaging the central 90 percent of the price change distribution each month. In other words, it drops the components with more extreme monthly price variations up to the point in which these sum up to the 10 percent of the distribution (5 percent of each tail). The second index is the asymmetrically trimmed mean. This is a variation of the previous index. The distribution is truncated in an asymmetric way in this index.57 The rationale for the trimmed mean indices is that extreme variations in prices are associated with supply shocks, while demand pressures are better captured by price movements coming from the central part of the density.58
The third group includes indices obtained through statistical filtering of the CPI. We consider three indices within this group. First, we analyze the 12-month and the 24-month lagged moving average of the CPI. As we argued before, the main motivation for analyzing the statistical properties of these indicators comes from the fact that, given their simplicity, most central banks that pursue inflation targeting policies focus on them. Finally, following Bryan and Cecchetti (2000), we also consider a 12-month moving average of the asymmetrically trimmed indicator. Then, after correcting for the extreme observations in the price distribution, we also apply time series smoothing to the index.
79. Figure 1 describes the evolution of the different CIIs, contrasting them with the evolution of the CPI.
Colombia Core Inflation Indicators and CPI
(Monthly annual inflation rates)
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
Colombia Core Inflation Indicators and CPI
(Monthly annual inflation rates)
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
Colombia Core Inflation Indicators and CPI
(Monthly annual inflation rates)
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
C. Evaluation of Core Inflation Indicators
80. In the introduction, we mentioned the essential statistical properties of CIIs. We analyze them in this section, comparing the properties of every CII with those of the consumer price index and other CIIs.
81. Central banks that pursue an inflation targeting policy seek to target the trend in the inflation rate determined by the evolution of monetary aggregates. Excessive noise (or transitory fluctuations) in the inflation rate due to the effects of supply shocks undermines the capacity of the authorities to assess and control the evolution of inflation in the medium and long run. This is because non-monetary shocks of known short duration may not be taken into account by price setters in long horizons59. For this reason, small variability is an important characteristic of suitable CIIs. Following Jaramillo et al. (1999), we calculate the standard deviation of the series in relation to their trend value.60 The latter is computed by the Hodrick-Prescott filter of the series. Table 1 shows the results. It is remarkable that the fixed trimmed weight indicator is one of the least volatile indices. In fact, it matches the variability of two of the indices constructed through statistical smoothing techniques. These latter indices, given the way they are constructed, will generally have the smallest variability.
Colombia: Variability of Core Inflation Indicators
(January 1990–July 2000)
Root-mean-squared-error of the relevant core inflation indicator with respect to its trend (given by the Hodrick-Prescott Filter)
Colombia: Variability of Core Inflation Indicators
(January 1990–July 2000)
| Indicator | R.M.S.E 1/ | |
|---|---|---|
| CPI | 1.63 | |
| CPI without | ||
| food | 1.29 | |
| Fixed trimmed | ||
| weights 20 | 0.97 | |
| Trimmed | ||
| mean 10 | 1.40 | |
| Asymmetrically | ||
| trimmed mean | 1.39 | |
| Moving | ||
| average 12 | 1.00 | |
| Moving | ||
| average 24 | 0.52 | |
| ATM | ||
| moving average 12 | 0.94 | |
Root-mean-squared-error of the relevant core inflation indicator with respect to its trend (given by the Hodrick-Prescott Filter)
Colombia: Variability of Core Inflation Indicators
(January 1990–July 2000)
| Indicator | R.M.S.E 1/ | |
|---|---|---|
| CPI | 1.63 | |
| CPI without | ||
| food | 1.29 | |
| Fixed trimmed | ||
| weights 20 | 0.97 | |
| Trimmed | ||
| mean 10 | 1.40 | |
| Asymmetrically | ||
| trimmed mean | 1.39 | |
| Moving | ||
| average 12 | 1.00 | |
| Moving | ||
| average 24 | 0.52 | |
| ATM | ||
| moving average 12 | 0.94 | |
Root-mean-squared-error of the relevant core inflation indicator with respect to its trend (given by the Hodrick-Prescott Filter)
82. Another important desirable characteristic of a CII is related to bias. Although the CPI can be very volatile in the short term, the ultimate objective of the central bank is to control its evolution. A target index should not be systematically above or below the value of the CPI, and over long horizons, the mean value of any CII should coincide with the one of the CPI. We analyze the behavior of the CIIs in Table 2. The table presents the mean value of the different price indices for the period January 1990–July 2000. We perform tests of equality of means for each CII and find that the hypothesis of different means is rejected in all cases. Thus, all CIIs considered in this note seem to replicate reasonably well the evolution of the CPI in the long run. We also calculate deviations of the CIIs with respect to the CPI during the whole sample period. We find that the indices that better traced the evolution of the CPI, period by period, are the trimmed mean indices.
Bias of Core Inflation Indicators
(January 1990–July 2000)
Tests of equality of means were performed for each core inflation indicator, the null hypothesis was never rejected.
Root-mean-squared-error of the relevant core inflation indicator with respect to CPI.
Bias of Core Inflation Indicators
(January 1990–July 2000)
| Deviation | ||||
|---|---|---|---|---|
| Indicator | Mean 1/ | R.M.S.E 2/ | Max | Min |
| CPI | 21.48 | 0.00 | 0.00 | 0.00 |
| CPI without Food | 22.30 | 2.06 | 3.96 | -5.89 |
| Fixed trimmed weights 20 | 20.88 | 1.40 | 3.45 | -3.00 |
| Trimmed mean 10 | 21.00 | 0.96 | 2.33 | -1.89 |
| Asymmetrically trimmed mean | 20.72 | 1.33 | 4.08 | -2.14 |
| Moving average 12 | 21.74 | 2.02 | 3.32 | -6.31 |
| Moving average 24 | 21.87 | 3.05 | 1.33 | -8.02 |
| ATM moving average 12 | 21.04 | 2.01 | 4.27 | -6.33 |
Tests of equality of means were performed for each core inflation indicator, the null hypothesis was never rejected.
Root-mean-squared-error of the relevant core inflation indicator with respect to CPI.
Bias of Core Inflation Indicators
(January 1990–July 2000)
| Deviation | ||||
|---|---|---|---|---|
| Indicator | Mean 1/ | R.M.S.E 2/ | Max | Min |
| CPI | 21.48 | 0.00 | 0.00 | 0.00 |
| CPI without Food | 22.30 | 2.06 | 3.96 | -5.89 |
| Fixed trimmed weights 20 | 20.88 | 1.40 | 3.45 | -3.00 |
| Trimmed mean 10 | 21.00 | 0.96 | 2.33 | -1.89 |
| Asymmetrically trimmed mean | 20.72 | 1.33 | 4.08 | -2.14 |
| Moving average 12 | 21.74 | 2.02 | 3.32 | -6.31 |
| Moving average 24 | 21.87 | 3.05 | 1.33 | -8.02 |
| ATM moving average 12 | 21.04 | 2.01 | 4.27 | -6.33 |
Tests of equality of means were performed for each core inflation indicator, the null hypothesis was never rejected.
Root-mean-squared-error of the relevant core inflation indicator with respect to CPI.
83. We will pursue now the analysis of the last two characteristics of the CIIs, related to their forecasting power and their relationship with the monetary aggregates. Before we address them, it is worth keeping in mind that the behavior of the inflation rate in Colombia may have changed during the 1990s. Monthly observations of annual inflation rates during the ten-year period 1988–98 averaged almost 24 percent. The average from the end of that period to the present is less than 11 percent.61 This fact is especially relevant for the analysis that follows. There is an obvious issue about stability of coefficients (or even functional relationships) when dealing with forecasting or with the relationship of the CIIs with the monetary aggregates. The analysis below should be interpreted as a first assessment of these issues, keeping in mind that a more involved analysis may be necessary for a thorough assessment of these problems.
84. Next, we set out to examine the ability of the different indices to forecast the evolution of the CPI for two different time horizons, 12 months and 24 months. We aim to assess how accurately lagged values of the different price indices can forecast the behavior of the CPI.62 Table 3 shows the results and explains in detail the estimation method. The table suggests that, with the exception of the two moving average indices, the CIIs present lower forecast errors than the CPI.63 In other words, the CPI does a poorer job than the other indices in forecasting itself. Limited influence indicators, such as the CPI without food and the fixed trimmed weights, have the lowest errors in 12-month forecasting. The 12-month moving average of the asymmetrically trimmed mean indicator (ATM12) has lower errors than any other index in 24-month forecasting. Overall, forecasting errors are low and stable with the ATM 12 index. This suggests that this index is the best predictor of the CPI.
Forecasting Power of Core Inflation Indicators 1/
(Complete Sample: January 1990–July 2000)
Method of estimation of forecast values is recursive least squares. A period of five years was used for the initial parameter estimation. These are the immediate previous years to the one-step forecast estimation dates reported in footnote 2.
Root-mean-squared-error of forecasts of inflation beginning in the following dates: (i) for CPI, CPIWF, NUC20, TM10 and ATM, the initial forecasting dates are 1996:01-2000:07 for the 12-month horizon and 1997:01-2000:07 for the 24-month horizon, (ii) for MA12 and ATM12 the dates are 1997:01-2000:07 for the 12-month horizon and 1998:01-2000:07 24-month horizon and (iii) for MA24 the dates are 1998:01-2000:07 and 1999:01-2000:07 for the same horizons.
Forecasting Power of Core Inflation Indicators 1/
(Complete Sample: January 1990–July 2000)
| Indicator | 12-month Horizon R.M.S.E 2/ | 24-month Horizon R.M.S.E 2/ |
|---|---|---|
| CPI | 4.82 | 6.71 |
| CPI without food (CPIWF) | 2.82 | 6.01 |
| Fixed trimmed weights 20 (NUC20) | 3.33 | 5.45 |
| Trimmed mean 10 (TM10) | 4.43 | 6.42 |
| Asymmetrically trimmed mean (ATM) | 4.01 | 6.11 |
| Moving average 12 (MA12) | 5.70 | 6.62 |
| Moving average 24 (MA24) | 6.77 | 8.00 |
| ATM Moving average 12 (ATM12) | 3.56 | 3.64 |
Method of estimation of forecast values is recursive least squares. A period of five years was used for the initial parameter estimation. These are the immediate previous years to the one-step forecast estimation dates reported in footnote 2.
Root-mean-squared-error of forecasts of inflation beginning in the following dates: (i) for CPI, CPIWF, NUC20, TM10 and ATM, the initial forecasting dates are 1996:01-2000:07 for the 12-month horizon and 1997:01-2000:07 for the 24-month horizon, (ii) for MA12 and ATM12 the dates are 1997:01-2000:07 for the 12-month horizon and 1998:01-2000:07 24-month horizon and (iii) for MA24 the dates are 1998:01-2000:07 and 1999:01-2000:07 for the same horizons.
Forecasting Power of Core Inflation Indicators 1/
(Complete Sample: January 1990–July 2000)
| Indicator | 12-month Horizon R.M.S.E 2/ | 24-month Horizon R.M.S.E 2/ |
|---|---|---|
| CPI | 4.82 | 6.71 |
| CPI without food (CPIWF) | 2.82 | 6.01 |
| Fixed trimmed weights 20 (NUC20) | 3.33 | 5.45 |
| Trimmed mean 10 (TM10) | 4.43 | 6.42 |
| Asymmetrically trimmed mean (ATM) | 4.01 | 6.11 |
| Moving average 12 (MA12) | 5.70 | 6.62 |
| Moving average 24 (MA24) | 6.77 | 8.00 |
| ATM Moving average 12 (ATM12) | 3.56 | 3.64 |
Method of estimation of forecast values is recursive least squares. A period of five years was used for the initial parameter estimation. These are the immediate previous years to the one-step forecast estimation dates reported in footnote 2.
Root-mean-squared-error of forecasts of inflation beginning in the following dates: (i) for CPI, CPIWF, NUC20, TM10 and ATM, the initial forecasting dates are 1996:01-2000:07 for the 12-month horizon and 1997:01-2000:07 for the 24-month horizon, (ii) for MA12 and ATM12 the dates are 1997:01-2000:07 for the 12-month horizon and 1998:01-2000:07 24-month horizon and (iii) for MA24 the dates are 1998:01-2000:07 and 1999:01-2000:07 for the same horizons.
85. The last characteristic is related to the relationship of the CIIs with the monetary aggregates. We consider two monetary aggregates in this note: (a) monetary base,64 and (b) Ml. Then, we analyze: (i) if there exists a long-term relationship between monetary aggregates and CIIs, (ii) if it is the behavior of monetary aggregates that influence the behavior of prices or the other way around and,65 (iii) if monetary aggregates can better explain the behavior of the CIIs than the CPI. We find that:
Cointegration tests66 suggest that there exists a long-term relationship between all CIIs and two monetary aggregates, monetary base and Ml.
The only CII that seems to be influenced by monetary aggregates without influencing them at the same time is the 12-month moving average index.67
The monetary base (MB) seems to have the highest explanatory power of the behavior of prices. The MB explains better all the CIIs than it does the CPI (see Table 4).
The behavior of the 24-month moving average index (MA24) and the ATM 12 are well traced by the evolution of the MB. The evidence suggests that the MA24 has the most stable relation with the MB.68
Core Inflation and Monetary Aggregates
(January 1990–July 2000)
Adjusted R2 from a regression of twenty-four lags of money growth on inflation.
Monetary base adjusted by changes in reserve requirements.
Monetary aggregates were found weakly exogenous to MA12, according to Johansen (1992) criteria.
Core Inflation and Monetary Aggregates
(January 1990–July 2000)
| Explanatory Power of Monetary Aggregates 1/ | ||
|---|---|---|
| Indicator | Adjusted Base 2/ | Ml |
| CPI | 0.67 | 0.41 |
| CPI without food (CPIWF) | 0.80 | 0.15 |
| Fixed trimmed weights 20 (NUC20) | 0.82 | 0.32 |
| Trimmed Mean 10 (TM10) | 0.75 | 0.41 |
| Asymmetrically trimmed mean (ATM) | 0.75 | 0.38 |
| Moving average 12 (MA12) 3/ | 0.81 | 0.41 |
| Moving average 24 (MA24) | 0.91 | 0.32 |
| ATM moving average 12 (ATM12) | 0.87 | 0.34 |
Adjusted R2 from a regression of twenty-four lags of money growth on inflation.
Monetary base adjusted by changes in reserve requirements.
Monetary aggregates were found weakly exogenous to MA12, according to Johansen (1992) criteria.
Core Inflation and Monetary Aggregates
(January 1990–July 2000)
| Explanatory Power of Monetary Aggregates 1/ | ||
|---|---|---|
| Indicator | Adjusted Base 2/ | Ml |
| CPI | 0.67 | 0.41 |
| CPI without food (CPIWF) | 0.80 | 0.15 |
| Fixed trimmed weights 20 (NUC20) | 0.82 | 0.32 |
| Trimmed Mean 10 (TM10) | 0.75 | 0.41 |
| Asymmetrically trimmed mean (ATM) | 0.75 | 0.38 |
| Moving average 12 (MA12) 3/ | 0.81 | 0.41 |
| Moving average 24 (MA24) | 0.91 | 0.32 |
| ATM moving average 12 (ATM12) | 0.87 | 0.34 |
Adjusted R2 from a regression of twenty-four lags of money growth on inflation.
Monetary base adjusted by changes in reserve requirements.
Monetary aggregates were found weakly exogenous to MA12, according to Johansen (1992) criteria.
D. Conclusion
86. This note analyzed the behavior of CIIs in Colombia during the 1990s. We use four different criteria to compare the CIIs with the CPI. These are: bias, volatility, forecasting power, and relation to monetary aggregates. We find that the CIIs trace the evolution of the CPI reasonably well, being at the same time less volatile indices. As for inflation forecasting, the CIIs have lower forecasting errors than the CPI. Exceptions to this are the statistical filters of the CPI, which have higher forecasting errors than the CPI itself. Finally, regarding the capacity of monetary aggregates to explain the behavior of prices, the data suggest that there is a long-term correlation between the behavior of money and the CIIs. Nonetheless, causality tests suggest the possibility of endogenous money. In other words, it seems that the behavior of prices explains the dynamics of the monetary aggregates.
87. By looking at the different criteria more closely, we can contrast the characteristics of the different CIIs. All indices trace fairly well the evolution of the CPI in the long run. By contrast, it is clear that in terms of volatility the statistical filters have the best performance among CIIs.69 Regarding forecasting power, the data suggest that limited influence indicators outperform statistical filters, but not the ATM12 index. The stability of the low forecasting errors for different forecasting horizons of this index is remarkable. Finally, results concerning money and prices suggest the existence of a causality problem70. Nevertheless, it is clear that the CIIs have stronger correlations with monetary aggregates than the CPI.
88. Overall, it seems that the combination of limited influence techniques and statistical smoothing seems to offer a suitable instrument to target inflation on the part of the monetary authority. The 12-month moving average of the ATM makes a good trace of the CPI in the long run, has small variability, and its forecasting errors are low and remarkably stable when considering different forecasting horizons. It may be of interest to the Colombian authorities to focus on the behavior of this index.
Colombia: Monetary Base and Core Inflation Parameter Stability Tests 1/
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
1/ Recursive Least Square Estimate of twenty-four lags of money growth on inflation (estimation dates as defined in Table 3) Cumulated Residual Test of Parameter Stability.Colombia: Monetary Base and Core Inflation Parameter Stability Tests 1/
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
1/ Recursive Least Square Estimate of twenty-four lags of money growth on inflation (estimation dates as defined in Table 3) Cumulated Residual Test of Parameter Stability.Colombia: Monetary Base and Core Inflation Parameter Stability Tests 1/
Citation: IMF Staff Country Reports 2001, 068; 10.5089/9781451808834.002.A005
1/ Recursive Least Square Estimate of twenty-four lags of money growth on inflation (estimation dates as defined in Table 3) Cumulated Residual Test of Parameter Stability.References
Bryan, M. and S. Cecchetti, 1994, “Measuring Core Inflation,” in Monetary Policy, edited by Gregory Mankiw, The University of Chicago Press.
Bryan, M. and S. Cecchetti, 1996, “Measuring Short-Run Inflation for Central Bankers,” WP 5786, National Bureau of Economic Research.
Bryan, M. and S. Cecchetti, 2000 “A Note on the Efficient Estimation of Inflation in Brazil,” Mimeo.
Debelle, G., P. Masson, M. Savastano and S. Sharma, 1998, “Inflation Targeting as a Framework for Monetary Policy,” Economic Issues of the IMF No. 15.
Drazen, Allan, 2000, “Political Economy in Macroeconomics,” Princeton University Press.
Jaramillo, Carlos, 1998, “Measuring Inflation Using Asymmetric Means,” Borradores de Economía del Banco de la República No. 91.
Jaramillo C., E. Caicedo, A. Cobo, A. González, M. Jalil, J. Julio and L. Melo, 1999, “La Inflación Básica en Colombia: Evaluación de Indicadores Alternativos,” Borradores de Economía del Banco de la República No. 136.
Rivas, L. and J. Rojas, “Precios Relativos, Inflación Subyacente y Metas de Inflación: Un Análisis para Nicaragua,” Banco Central de Nicaragua, Gerencia de Estudios Económicos.
Prepared by Esteban Vesperoni.
Among them, we find New Zealand, Canada, the United Kingdom, Finland, Sweden, Australia, and Spain.
See, for example, Drazen (2000) and Debelle et. al. (1998).
See, for example, Bryan and Cecchetti (1994, 1996, and 2000) and Jaramillo et al. (1999).
Most central banks with explicit inflation targeting policies focus on 12-month trends of inflation, given the simplicity of these indicators.
A statistical filter is a procedure to smooth out the behavior of the series, like moving averages or Hodrick-Prescott filters. These techniques isolate the trend of the series from short-term volatility.
Many food products have very elastic supplies.
The period considered for this index was January 1990–April 1999.
It intends to take into account that the distribution of price changes in Colombia is asymmetric, in the sense that it tends to concentrate values in one of the two tails (skewness is different from zero in these distributions). For a case like this, it can be more accurate to build indicators which are truncated asymmetrically. In this index, the distribution was truncated at the 15 percentile in the left tail and at the 13 percentile in the right tail (see Jaramillo (1998) and Jaramillo and Córdoba (1997)).
Technically, Bryan and Cecchetti (2000) show that this problem is associated with exceptionally leptokurtic distributions, that is, with distributions with “fat tails.” In these cases, measuring price changes through weighted means may not be efficient.
Given that the policy instruments of central banks have a considerable lag to affect economic activity, the effects of supply shocks may have vanished by the time monetary policy is effective.
As Jaramillo et al. state, this strategy is due to the fact that price variation series in Colombia are nonstationary and present a clearly declining trend during the 1990s.
While the annual inflation rate in January 1990 was almost 27 percent, the same value by the end of 2000 was 8.75 percent.
We run recursive least squares to evaluate this, using initial estimation periods of five years and then performing one-step-forecasts with the rest of the sample.
Forecast errors are the usual criteria to compare predictive power of different variables or model specifications. These are calculated as the root-mean-squared-error (RMSE) of the forecast of inflation. First, a relation between the variable under consideration and the CPI is estimated. Then, forecast periods are chosen (see Table 3) and the estimated relation is used in order to predict the value of the CPI. Then, this predicted value is compared to the actual value of the series to obtain the RMSE. Lower RMSE imply better predictors of the CPI.
Adjusted by changes in reserve requirements.
The possibility of endogenous money, that is, situations in which the evolution of monetary aggregates could be driven, at least partially, by the behavior of prices. This is especially relevant in high inflation economies, where either due to institutional arrangements (like formal backward indexation of contracts) or to expectations (inflation inertia), price setting attains a dynamics of its own. The reaction of monetary aggregates to the price dynamics depends on the behavior of private agents and the central bank. If one focuses on broad monetary aggregates, private agents can create money by changing their portfolio of financial assets or their trading practices (inside money and velocity). Regarding the monetary authorities, the behavior of base money will depend on the objectives of the central bank. If the latter is concerned about economic activity, price dynamics may force the authorities to corroborate price increases (passive money) or face a possible economic downturn for not doing so.
That is, tests that assess if there is comovement between variables in the long run.
Johansen’s criterion of weak exogeneity was used.
This is hardly surprising, given that the statistical filter works by smoothing out the price series. Notice, though, the good performance of the fixed trimmed weights indicator in this regard (this is consistent with Jaramillo et al. (1999)).
And, hence, the need to address the problem with different econometric techniques.
