Options-based forecasts of futures prices in the presence of limit moves

Abstract

The reported analysis examines a simultaneous estimation option-based approach to forecast futures prices in the presence of daily price limit moves. The procedure explicitly allows for changing implied volatilities by estimating the implied futures price and the implied volatility simultaneously. Using futures and futures options data for three agricultural commodities, it is found that the simultaneous estimation approach accounts for the abrupt changes in implied volatility associated with limit moves and generates more accurate price forecasts than conventional methods that rely on only one implied variable.

Notes

¹ While potentially important, examination of the data demonstrated little evidence of a pronounced smile for the three commodities investigated here.

² Crop production reports are released at 8:30 am EST (3:00 pm EST before 1995) and hogs and pigs reports are released at 3:00 pm EST by the National Agricultural Statistics Service (NASS), U.S. Department of Agriculture.

³ A reviewer suggested we compare the simultaneous procedure to the last observed pre-limit futures price. In all cases, the last pre-limit futures price possessed poor forecasting ability, significantly worse than the alternatives examined here.

⁴ To further explore the effects of seasonality on forecast ability for corn and soybeans, we performed the analysis on the observations that fell during the periods of high volatility. Our findings are robust and are consistent with the properties of the MDM test.

⁵ A reviewer asked us to consider the presence of an overnight effect which is commonly measured in time series models estimated using intra-day data over a number of days. The overnight effect measures the regular change in price due to the fact that first observation of a new day may contain more information. In our context, the issue is less meaningful as a limit move results in ex ante uncertainty on both the price level and when the market will resume trading. Nevertheless, we investigated the issue by splitting samples into observations where the futures market resumed trading on the day of the limit move and where the trading begins after the day of the limit move. As expected, larger forecast errors almost always exist for observations where trading does not resume on the day of the limit move. With regards to the relative performance of the procedures, the SEA continues to perform well but the PIF performs modestly better for soybeans at the longer temporal horizon, possibly due to the smaller number of observations (8) and larger errors.