Uncertainty and overconfidence in time series forecasts: application to the Standard & Poor's 500 Stock Index
Authors: Gordon D. A.; Kammen D. M.
Source: Applied Financial Economics, Volume 6, Number 3, 1 June 1996 , pp. 189-198(10)
Key:
- Free Content
- New Content
- Subscribed Content
- Free Trial Content
- Free Content
- New Content
- Subscribed Content
- Free Trial Content
Abstract:
Both day-to-day and long-term decisions often rely on some form of forecasting, where misspecification of the probability of some future value can mean misapplication of policies or misdirected business decisions. Many forecasts explicitly or implicitly assume a Gaussian confidence interval; but the assumption of a normal distribution for forecast errors may systematically underrepresent outcomes for data far from the mean. The degree of overconfidence underlying Gaussian intervals has previously been evaluated and an ex post methodology proposed for correcting for forecast uncertainty. This present study applies the same methodology to financial forecasts of returns of the Standard and Poor's (S&P) Composite Index, projections which follow a Markov process. It is found that Gaussian-style forecasts of S&P returns exhibit systematic 'overconfidence', requiring an empirical inflation factor of 3.44 in order to more accurately represent the stated level of confidence. These results suggest that even comparatively 'well calibrated' short-term (one day) forecasts may exhibit mismeasured uncertainty, impacting forecast utility.Language: English
Document Type: Research article
Key:
- Free Content
- New Content
- Subscribed Content
- Free Trial Content
- Free Content
- New Content
- Subscribed Content
- Free Trial Content
Click here for Page Help