Title: Commodity price shocks and the abnormal returns of stocks: Evidence from competing dynamic linear models Speaker: David C. BROADSTOCK, Hong Kong Polytechnic University Host: Jingjing Ye, Associate professor, RIEM Time: 14:00-16:30, October 14, Friday Venue: Yide building H503, Liulin Campus Abstract: This study is motivated by a growing interest among energy economists, and empirical econometricians more widely, to model empirical economic relationships in such a manner that they may evolve over time. More precisely the class of empirical techniques which enable time-varying parameters to be obtained. Both approximate (recursive OLS) and explicit methodologies (Kalman Filter, KF) are available to the researcher. The approximate approach is appealing owing to computation simplicity, while the KF is somewhat more demanding yet theoretically superior in many ways. The aim of this work is to explore under what conditions an approximate approach may be either equally, or better performing than the KF. The approach to the study involves two phases, one to establish the statistical significance of the difference between the approaches (in terms of their ability to accurately describe some known data generating processes), the other to establish the economic significance of differences in results/policy implications that arise from using the different methods. A Monte Carlo experiment is carefully structured to create a range of data generating processes that are deliberately tailored to enable the `properties' of the approximate approach to `shine', and hamper the theoretical accuracy of KF, though environments suited toward the KF are included for completeness. Early results provide a compelling case that the KF could be a more accurate modelling solution across a wide range of environments, including those where it should in principle struggle to perform well. To hint towards the economic significance of model choice, each of the models are extended towards a simple time-varying risk prediction problem. The empirical component is designed to be of standalone interest, exploring the extent to which stock prices can be seen to react to commodity price uncertainty. The notion of risk in question is an approximation to historical volatility (connecting to the second moment of stock returns) based on expected intra-day price variation, calculated for more than 250 separate stocks from the Hong Kong stock market from 1990 until today. The empirical question(s) being then do (non-parametrically defined as per Dimitropoulos and Yatchew, forthcoming) shocks in commodity prices stimulate abnormal levels of volatility? How long do these effects persist, what sign do they take, and might the scale of effect be connected in some way to the elasticity of demand for the commodity? In summary the contributions of this paper are to highlight not only the potential limitations of approximating time-vary effects with seemingly intuitive and appealing alternatives. The study also serves to illustrate the strength of explicitly time-varying models to analyze data, across a wide range of scenarios, including those where the methods might a-priori be expected to be inferior. It is additionally demonstrated that stock market volatility can be predicted by off-trend spikes in commodity prices, especially when the price spikes are large and the underlying commodity price lies towards the upper tail of the historical price distribution. Keywords: Time-varying parameters; Kalman filter; rolling window regressions; Monte-Carlo.