A0695
Title: Predicting pull and bear markets: On pooling opinions and sharpness
Authors: Shui Ki Wan - Hong Kong Baptist University (Hong Kong) [presenting]
Abstract: Using monthly S\&P 500 data, we assess the predictability of 20 financial and macroeconomic indicators for the bulls and bears. The time-varying likelihood ratio statistics suggest that some of the indicators do have in-sample predictability but are highly volatile. Benchmarking on the historical average, the out-of-sample statistic quadratic probability scores (QPS) suggest that the static dividend-price ratio, stock variance, non-farm payroll, federal fund rates and recession probabilities models are highly predictive for all the trend definitions. The dynamic model further improves the forecasting performance significantly. Their improvement is mainly attributed to the substantial increment in sharpness. Since investors are likely to be presented with a pool of forecasts, we also consider linear combinations of the probability forecasts. Although they lead to uncalibrated forecasts that are lack of sharpness, we find that the pooled forecasts are still well-calibrated with the sharpness that are close to the best model. Our studies also show that the adaptive opinion pooling using out-of-sample performance is better than other weighting scheme based on in-sample statistics. By putting sharpness into perspective, we find that trading strategy utilizing sharpness can generate a higher portfolio returns than the traditional threshold approach.
