Modelling inflation in Georgia

Abstract

The article explains the behaviour of inflation in Georgia in the post-stabilization period. Long-run equation linking prices with money and exchange rate, as well as short-run, dynamic equation for inflation are estimated. The inflation equation is stable, points to a dominant role of exchange rate in the inflation behaviour. The equation explains well the behaviour of inflation after the Russian crises, when inflation increased sharply but was quickly brought under control when the National Bank of Georgia kept its monetary policy tight and exchange rate stable.

Notes

¹Measured as a ratio of foreign to total deposits.

²Another implication is that, even if it is possible to successfully find one cointegrating vector corresponding to Equation 1, a stable money demand function cannot be estimated from the data.

³Taylor (1991) proposed a methodology to handle inflationary expectations in the cointegrated money demand framework under rational expectations. See Frenkel and Taylor (1993), or more recently Choudhry (1998) and Budina et al. (2006), for applications.

⁴The code for the Kalman filter interpolation – written in Ox (Doornik, 1999) with the SSFPack package (Koopman et al ., 1999) – is available on request from the author.

⁵Estimation of the model using M3 instead of M2 yields very similar results.

⁶As implemented in GiveWin. Ericsson et al . (1994) show that the linear seasonal filter approximating earlier version of the Census procedure (X-11) may affect inferences about cointegration and alter dynamics and exogeneity status of variables in the model. Siklos et al . (1995) show that the X-11 procedure exhibits strong nonlinear characteristics and that these characteristics may lead to incorrect inference about the number of cointegrating vectors in series simulated from linear models. However, the exchange rate dynamics in Georgia exhibits a pronounced seasonal component in the post-1998-depreciation period, which does not exist in the pre-crisis period. Accounting for this effect in a model with seasonally unadjusted data would be difficult. Seasonally adjusting the exchange rate series for the post-crisis period only (and other variables for the whole period) is a more straightforward strategy, which is applied in the article. More generally, empirical series may exhibit certain nonlinear features and the impact of adjustment procedures on more complicated process has not yet been fully researched.

⁷The log of real GDP is interpolated under the assumption that the series follows a unit root process.

⁸Sequential testing starting from the highest order of six allows for reduction of the lag length to four. Given high uncertainty surrounding the correct lag length I opt for the over-parameterized model. Monte Carlo studies in Gonzalo (1994) show that efficiency loss from choosing a too long-lag structure is small, while a too short-lag structure has a severe impact on maximum likelihood estimates. Estimates of cointegrating vector obtained from the four-lag model are almost identical to those obtained from the six-lag specification.

⁹Estimating the model with a trend added to the cointegrating space does not change the number of cointegrating vectors, but results in very high SEs of the trend and real output coefficients.

¹⁰ F-tests have been used to test the restrictions. Akaike, Schwarz, Hannan-Quinn and the final prediction error criteria have been used for judging adequacy of the reductions.

¹¹SEs of some coefficients are higher, but this is not surprising, given that the new estimates are based on a shorter and therefore less informative sample.

¹²The estimated long-run relationship contains the same variables (prices, money, exchange rate and output) as some money demand studies for developing and emerging countries (e.g. Bahmani-Oskooeea and Rehmanb, 2005, Bjornland, 2005), but the theoretical model in this article implies different coefficient restrictions. In particular, the model implies a positive coefficient for the level of exchange rate, which is the opposite to the portfolio balance model of money demand (Arango and Nadiri, 1981). Some studies report money demand functions with a positive exchange rate coefficient in the above cointegrating relation – inconsistent with the portfolio balance model – and argue that the level of exchange rate is a proxy for depreciation expectations (Biornland, 2005). The coefficient of the exchange rate seems easier to interpret in the model presented in this article, as exchange rate depreciation is a more natural proxy for depreciation expectations (e.g. Phylaktis and Taylor, 1993; Slavova 2003; Adam et al ., 2004).