Forecasting Chilean inflation with international factors

West (2008) is a presentation based on the article by Engel et al. (2012).

In our sample of OECD economies, we rule out the cases of New Zealand and Australia due to the unavailability of CPI information at a monthly frequency. We also rule out the case of Estonia due to data availability. For Estonia we only find data for the year 1998 onward. To reduce distortions coming from considerations of different sample periods, we just work with the list of 30 OECD economies with information at a monthly frequency during the entire sample period January 1994–March 2013. Finally, from the group of LATAM countries we remove Cuba because we were not able to find official CPI data at a monthly frequency.

We have a very simple justification for the choice of year-on-year inflation as a target variable: To our knowledge, every inflation targeting country in the world defines its target in year-on-year terms. For instance, the Czech Republic has a target of 2 % for the medium term. The UK has the same target but is supposed to be met at all times. In Thailand and Mexico, the target is 3 %. Some countries have a target of 2.5 % such as Iceland, Norway, Poland, Romania and North Korea. The list is long but all of these countries express their target in year-on-year terms. In particular, Chile has a target of 3 % and a tolerance band between 2 and 4 %. According to Chilean monetary authorities, year-on-year inflation in Chile is expected to lie within this band “most of the time”. Given that monetary authorities have defined their target in year-on-year terms, we think that forecasting year-on-year inflation is a reasonable thing to do.

We notice that in this paper we use the unconditional and univariate version of the GW test that follows an asymptotic normal distribution. In Pincheira (2013) there is a detailed discussion about the linkage between the Clark and West (2007) test and tests based on unadjusted comparisons of MSPE for the particular case in which the null hypothesis is a martingale difference model.

For instance, in the case \(c=0.02\) and \(\alpha =0.95\) the slope of the linear trend is 0.2. When forecasting 2 years ahead, \(h=24\), therefore the contribution of the deterministic term in (15) to the overall inflation forecast is 9.6 %. Furthermore, this contribution increases with the forecasting horizon.

The only exception is the case of the international inflation factor constructed with OECD countries when the Phillips-Perron test is used.

We also considered, as a robustness check, another exogenous variable: a Latin American exchange rate factor (LAERF). We computed this factor as the first principal component of the set of year-on-year variation of local exchange rates against the American dollar for 14 Latin American countries. We used monthly data from Bloomberg for the following countries: Argentina, Bolivia, Brazil, Colombia, Costa Rica, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru and Uruguay. This is basically the same sample of Latin American countries we used for the construction of the international inflation factor, but excluding Ecuador, Haiti, Dominican Republic and Venezuela. These countries were not considered due to missing observations in our database or because they had fixed exchange rates in long periods of our sample and consequently, in many of our out-of-sample estimation windows. Both in-sample and out-of-sample results with this additional exogenous variable are fairly similar to the results reported here. Accordingly, and for the sake of brevity, we do not report these additional results, but they are available upon request.

A referee pointed out that 24 lags “...seems an extremely high value”. To check this statement we also allowed \({q}_{ h} \) to take only the values 1, 2, 3 and 12 when computing our forecasts according to our specifications (17) and (19). Qualitatively our results were fairly similar. This is an important observation for policy makers because the computational time could be substantially reduced by considering a smaller set of lags for \({q}_{ h} \).

Chile is an OECD country since May 2010.