The ability to produce high-quality inflation forecasts is crucial for modern central banks. Inflation forecasts are needed for understanding current and forthcoming inflation trends, evaluating the effectiveness of previous policy actions, making new policy decisions, and building the credibility of a central bank in the eyes of the public. This motivates a constant search for new approaches to producing inflation forecasts. This paper analyses the empirical performance of several alternative inflation forecasting models based on structural vs. data-driven approaches, as well as aggregated vs. disaggregated data. It demonstrates that a combined ARMA model with data-based dummies that uses the disaggregated core inflation data for Ukraine allows to considerably improve the quality of an inflation forecast as compared to the core structural model based on aggregated data.

This paper develops an early warning model (EWM) for a micro-macro analysis of individual and aggregated bank vulnerabilities in Ukraine. We applied a stepwise logit for predicting defaults at Ukrainian banks based on a panel bank and macro-level data from Q1 2009 to Q3 2019. Next, we aggregated individual bank default probabilities to provide policymakers with information about the general state of the financial system with a particular focus on generating a signal for countercyclical capital buffer (CCB) activation. Our key findings suggest that the probability of default exceeding 11% could signal about a vulnerable state in a bank and, in the aggregated model, in a financial system in general. The aggregated model successfully issues an out-of-sample signal of a systemic crisis four periods ahead of the start of the 2014-2015 turmoil.

No abstract

This dissertation examines the empirical performance of several complete and incomplete market models of stock price dynamics using S&P 500 options and stock market data. The main contribution of this work is that it suggests and implementsg an empirical approach to estimating a complete model with uncertain volatility, and then judges it against other popular option pricing processes. The performance of alternative models is evaluated from four perspectives: (1) in-sample …t to stock returns data, (2) in-sample …t to options data, (3) consistency of physical and risk-neutral parameter estimates and (4) out-of-sample option pricing. Overall, the complete model with uncertain volatility is found to …t the data much better than models with constant and price-level-dependent volatilities, and the variance gamma process, and its performance is comparable to that of a stochastic volatility model.