Parametric early warning system model for Ukraine

There have been several crises in the world economy since the end of the last century. The developing economies were ones that have suffered the most, considering the level of openness, the weak institutional framework, and the market vulnerability to unpredictable shocks. One of the instruments widely used to prevent crises is Early Warning System models. The paper pursues a goal to develop a parametric logit/probit regression to determine early warning arguments and their appropriate thresholds for Ukraine. The logit/probit modeling corresponds to the determination of the dependent binary variable associated with an output gap followed by the selection of independent early warning arguments. The quarterly distribution of GDP transformed into monthly data by applying the Chow-Lin regression method of interpolating higher frequency values. The monthly GDP data employed for the evaluation of the output gap using a multivariate filter and Okun’s law definition. The average elasticity of change in the unemployment rate to GDP was 3. The 2% difference between the actual and potential GDP applied for generating binary data of the output gap. There were three independent variables favored to be early warning arguments of the logit/probit regression: supply-demand gap, the world price of raw materials, and broad money supply. The obtained econometric characteristics of the probit regression were more statistically significant in comparison to the logit model. There were similar culminating positions of the fitted indicator during the crises in 2008-2009 and 2014-2015 that were different in terms of the sources. The corresponding marginal effects for the supply-demand gap, the world price of raw materials, and the broad money supply were respectively 4.0%, 0.7%, and 0.4%. Regarding the higher marginal grade, the demand-supply gap is more significant among the given early warning components for predicting crises in Ukraine. In the following study, the new arguments of the logit/probit regression have to be examined to ...