Forecasting Macroeconomic Variables

Jour Fixe talk by Evi Salamaliki on February 5, 2015

What is the causal relationship between female labor supply and fertility in the US? Does economic policy uncertainty play an important role in macroeconomic performance? These are some  of the questions Evi Salamaliki deals with in her research and which she presented in her talk on “Trending macroeconomic time series and Granger causality”.

Her work is focused on time series data. A time series is defined as a sequence of values that a specific variable has taken on over some period of time. The observations have a natural ordering in time, the observation frequency can be monthly (high frequency), quarterly, or annual (low frequency).

Examples of economic time series are unemployment rates or a country´s real Gross Domestic Product. The main objectives of analyzing economic time series are forecasting: information about the likely future evolution of economic variables; and the dynamic interrelationships between a number of variables. To analyze these interrelationships the economist applies the Granger causality concept: “The analysis of Granger (1969) causality is one of the most popular tools in studying the dynamic interrelationships between sets of theory related economic variables in a VAR (vector autoregressive) framework”, she pointed out. “The standard notion of Granger causality restricts prediction improvement to a forecast horizon of one period, while it considers only direct flows of information between the variables of interest (direct causality). In VAR models with more than two variables, the standard Granger causality concept can be extended by examining prediction improvement at higher forecast horizons, so that indirect flows of information might be revealed through the additional system variables. This concept of causality is also referred to as indirect or multiple horizon causality.”

Evi Salamaliki´s main research objective is to investigate how the predictive ability changes under different trend treatments in VAR models. More precisely, she focuses on the cases  of structural shifts in the level and slope of the trend function (which, for series in logarithms, represent average growth rate changes). Generally, as to trends, the relevant time series literature distinguishes between stochastic trends (shocks induce permanent and non-vanishing effects), deterministic trends (shocks induce transitory effects or effects that die out quickly), and the presence of structural breaks/shifts in the form of infrequent changes in the deterministic trend (or a combination of the aforementioned cases). “Our research is based on simulations under different combinations of breaks and dynamic parameters, as well as on empirical investigations based on  specific VAR models.”