Narrow Results

In this research, monthly forward exchange rates are evaluated for possible existence of time varying risk premia in Singapore forward foreign exchange rates against US dollar. The time varying risk premia in Singapore dollar is modeled using non-Gaussian signal plus noise models that encompass non-normality and time varying volatility.

The results from signal plus noise models show statistically significant evidence of time varying risk premium in Singapore forward exchange rates although we failed to reject the hypotheses of no risk premium in the series. The results from Gaussian versions of these models are not much different and are in line with Wolff (1987) who also used the same methodology in Gaussian settings.

Our results show statistically significant evidence of volatility clustering in Singapore forward exchange rates. The results from Gaussian signal plus noise models also show statistically significant evidence of volatility clustering and non-normality in Singapore forward foreign exchange rates. Additional tests on the series show that exclusion of conditional heteroskedasticity from the signal plus noise models leads to false statistical inferences.

The main purpose of this article is to empirically demonstrate the effects of temporal aggregation when applying reverse regression models hypothesizing that spot prices today help predict forward rates in the future. This paper essentially reviews results from earlier research indicating that time-series aggregation will most certainly influence standard errors on parameter estimates. Standard errors are likely to increase with aggregation. The relationships between futures prices and spot oil prices are analyzed along with the importance of the effects of temporal aggregation and alternative model specification for understanding empirical relationships between the markets. Model specification and time-series aggregation over daily, weekly, and monthly aggregations confirm evidence that estimated standard errors are likely to increase with aggregation and t-ratios change as well. While goodness-of-fit measures might increase with aggregation, forecast accuracy with macrolevel aggregation might deteriorate owing to information loss due to the averaging of observations associated with underlying microstructures.

Statistical inference on the parameters of regression models requires special precautions when the error term is heteroskedastic and/or non-normal. In this case, although conventional test statistics do not follow t and F distributions, simulation methods can be used to draw inferences. We discuss two methods: the wild bootstrap and the generalised lambda distribution. By employing both artificial and real-world data from the National Footbal League, we show that these methods may prove particularly useful in hypothesis testing.

Booms and busts in economic activity are a regular occurrence. They lead to a strong empirical regularity, i.e. that output gaps and output growth are non-normally distributed. Mainstream macroeconomic models explain this phenomenon by invoking exogenous shocks that are non-normally distributed. This is not a very satisfactory explanation as it shifts our ignorance one step further. I propose an explanation based on a behavioral macroeconomic model, in which agents are assumed to have limited cognitive abilities and thus develop different beliefs. This model produces waves of optimism and pessimism in an endogenous way (animal spirits) and provides for a better (endogenous) explanation of the observed non-normality in output movements. I also analyze the implications for monetary policy.