Model specification and forecast accuracy
Forecasters and policy decision-makers often have to choose a model to use from a whole range of different models, all claiming to represent the economy (or the part of it that is the focal point of the forecasting exercise). The current range of wage and price models that can be used for inflation forecasting provides an example. As we have discussed earlier, in Chapter 9, inflation targeting implies that the central bank’s conditional forecast 1-2 years ahead becomes the intermediate target of monetary policy. Consequently, there is a strong linkage between model choice, forecasting, and policy analysis in this case.
The statistical foundation for a conditional forecast as an operational target is that forecasts calculated as the conditional mean are unbiased and no other predictor (conditional on the same information set) has smaller MSFE, provided the first two moments exist. However, as discussed earlier in this chapter, the practical relevance of the result is reduced by the implicit assumption that the model corresponds to the data generating process (DGP), and that the DGP is constant over the forecast horizon. Credible forecasting methods must take into account that neither condition is likely to be fulfilled in reality. However, the specific inflation models have one important trait in common: they explain inflation—a growth rate—by not only other growth rates but also cointegrating combinations of levels variables. Thus, they are explicitly or implicitly EqCMs.
Implications for inflation models’ forecasting properties between models with and without equilibrium-correcting mechanisms have been analysed in Bardsen et al. (2002a). This chapter draws on their results, and extends the analysis of different inflation models reported in earlier chapters. Specifically, we consider the two most popular inflation models, namely Phillips curves and dynamic wage curve specifications (or dynamic ICMs). These models were dealt with extensively in Chapters 4-6. The standard Phillips Curve Model (denoted PCM), is formally an EqCM, the cointegrating term being the output gap or, alternatively, the difference between the rate of unemployment and the natural rate, that is, ut — uphl1 in the notation of Chapter 4. An alternative to the PCM, consistent with the concept of a wage curve, was discussed extensively in Chapters 5 and 6, where it was dubbed the Imperfect Competition Model (ICM) because of the role played by bargaining and of imperfect competition. Since wage-curve models are EqCM specifications, they are vulnerable to regime shifts, for example, changes in equilibrium means.
The existing empirical evidence is mixed. Although varieties of Phillips curves appear to hold their ground when tested on United States data—see Fuhrer (1995), Gordon (1997), Blanchard and Katz (1999), Gall and Gertler (1999)—studies from Europe usually conclude that ICMs are preferable, see, for example, Dreze and Bean (1990, table 1.4), Wallis (1993), OECD (19976, table 1.A.1), and Nymoen and Rpdseth (2003). In Chapters 4 and 6 we presented both models (PCM and ICM) for Norway. In those models, and unlike most of the papers cited above, which focus only on wage formation or inflation, the rate of unemployment was modelled as part of the system. It transpired that the speed of adjustment in the PCM was so slow that little practical relevance could be attached to the formal dynamic stability of the PCM. No such inconsistency existed for the ICM, where the adjustment speed was fast, supporting strong dynamic stability.
Inflation targeting central banks seem to prefer the PCM, because it represents the consensus model, and it provides a simple way of incorporating the thesis about no long-run tradeoff between inflation and the activity level, which is seen as the backbone of inflation targeting (see, for example, King 1998).
In Section 11.3.1, we discuss the algebra of inflation forecasts based on the competing models. Section 11.3.3 evaluates the forecasting properties of the two models for Norwegian inflation.