Designing a Financial Condition Index to Predict Macroeconomic Variables Using Dynamic Time-varying Models

Objective: Since financial developments that are not driven by fundamental factors of monetary policy may have a large impact on the economy, policymakers have placed significant emphasis on the financial conditions index over the past recent years. Policymakers must maintain a vigilant watch over financial conditions, as their significance becomes increasingly significant. The construction and use of the financial condition index include three issues, which are: a) the selection of financial variables to enter the financial index, b) weights used to relate financial variables to the index, and finally c) the relationship between this index and the macroeconomy. There are many reasons for the changeability of these three cases over time, which can be discussed about the reasons for their occurrence and effect on the results. Many changes affect the way a financial index is made. Therefore, in this research, the goal is to design an index of financial conditions using time-varying dynamic models to improve the forecasting of macroeconomic variables.
Methods: The process of implementing the conceptual model can be explained in the following steps. First, extracting the desired variables to be used in the desired models (for this purpose, we have used monthly data during the years 2001 to 2021 for 19 financial variables and 5 macroeconomic variables). It should be noted that since all the variables must be in the form of rates - if necessary, all the variables were converted into growth rates. Also, all the variables were examined from the dimension of stationary, and the problem of their stationary has been solved. Second, the desired models were calculated to predict macroeconomic variables. Using time-varying models, dynamic averaging models, and dynamic selection models, the financial condition index was constructed in such a way that this index could include various variables to adopt different coefficients from its previous and subsequent periods and challenge constant parameter and variable models. Third, the predictive power of each of the proposed models in the estimation of macroeconomic variables (sum of squares of prediction error) was analyzed and estimated. Fourth, financial index extraction was done based on the model selected in the third step.
Results: The findings suggest that employing models that solely consider variability in the coefficients (without accounting for variability in the variables) leads to enhanced predictions of the unemployment rate compared to vector autoregression models and vector autoregression models incorporating the basic component. In addition, by moving from the generalized factor vector autoregression models and the time-varying generalized factor vector autoregression and the generalized factor time-varying vector autoregression, in which only the variability in the parameters is included (without considering the variability in the variables) towards the models that consider the variability in the parameters both in a Bayesian and dynamic way, a reduction will occur in the forecast error in the variables of the consumer price index, liquidity, monetary base, and gross domestic product. However, the amount of improvement in each of the variables is different from each other.
Conclusion: This study employed time-varying models to derive the financial condition index, aiming to provide the most accurate estimation of macroeconomic variables. The findings demonstrate that the utilization of these models outperforms other approaches in forecasting a majority of the macroeconomic variables