Modelling an Emergent Economy and Parameter Instability Problem

The paper examines a largely debated and important modelling problem – the instability of the econometric parameters, which has unpleasant consequences on both analytical and predictive planes. The modification of the estimators and/or of the standard deviations of a given model parameter when identical specifications and computational algorithms are applied on different statistical samples is considered to be its main manifestation. Several sources generating such a phenomenon may be identified: (i) statistical properties of the initial sample (mean, skewness, kurtosis, variance, outliers); (ii) econometric specification (selected explanatory factors and their functional connection with the dependent variables); (iii) changes in statistical data due to revisions; and (iv) dynamic instability, resulted from inherent changeability of real economic behaviours (consumer preferences, risk aversion of investors, households saving propensity, input-output coefficients and sectoral inter-flows, taxation system, domestic institutional context, foreign trade, international financial markets, and many other similar circumstances). For all these types, specific indicators and quantifying methodologies are proposed. As an illustration, we analysed the last version (2012) of the Romanian model, the macroeconomic parameters of which (182) were estimated using two computational algorithms (OLS and 2SLS) for seven samples: initial data for 1990-2011, the same series updated, and subsequent from 1990 to 2012, 2013, 2014, 2015, and 2016. Possible solutions to attenuate the parameter instability implications are also described, exemplifying them on the Romanian macromodel predictive simulations for 2017-2018 years. Some practical recommendations and further research topics are commented in the final part of the paper.