Prequential analysis of economic time series /
| Main Author: | |
|---|---|
| Other Authors: | , , |
| Format: | Thesis Book |
| Language: | English |
| Published: |
1991.
|
| Subjects: | |
| Online Access: | Link to OAKTrust copy |
| Abstract: | Prequential analysis of economic time series is applied to three cotton price series: world market price (AINDEX), the nearby futures price (NEARFT), and Memphis cash market price (PRCASH). Evaluation tools include: the probability score (PS), the ranked probability score (RPS), a sharpness measure, a measure of global calibration, and a chi-squared goodness of fit statistic. Previous applications of prequential analysis have stopped at the chi-squared goodness of fit statistic. Properties of these measures are studied. For "ordered" variables, the ranked probability score is suggested as the measure of probability forecasting "goodness". The cotton data series are divided into three parts. The first portion is used to estimate four different models: a univariate model (UNIV), an unrestricted VAR (UVAR), a restricted VAR (RVAR) and a Bayesian VAR (BVAR). The second portion of data is used to make and evaluate one, two, three, four, five and twenty-step-ahead probability forecasts. A bootstrap-like procedure is employed for this purpose. Uncertainty is permitted in both coefficients and residuals. The PS, RPS, SHP, GSB and chi-squared statistics are calculated using the issued probabilities and actual outcomes. Root mean-squared errors (RMSE) are calculated using the mean forecasts as the point forecasts. The results show that UNIV and RVAR models perform better than UVAR and BVAR model. At all horizons forecasts pass the calibration test, except at twenty-step horizon. Calibration functions indicate that variabilities of short-run forecasts are overestimated and variabilities of long-run forecasts are underestimated. Recalibration is considered for the restricted VAR model only. The third portion of the series is used to issue new set of forecasts which are recalibrated based on miscalibration in the second period. The results show that the recalibrated forecasts are not as good as the nonrecalibrated forecasts in accuracy, although sharpness has improved. Prequential analysis is applied to turning point prediction and conditional forecasting. The univariate model and the restricted VAR are used to forecast the next turning point, either a peak or trough... |
|---|---|
| Item Description: | Typescript (photocopy). Vita. "Major subject: Agricultural Economics." |
| Physical Description: | xi, 142 leaves : illustrations ; 29 cm |
| Bibliography: | Includes bibliographical references. |