Soybean Futures Price Forecasting Using Dynamic Model Averaging: Do the Predictors Change over Time?

ABSTRACT

This study uses the recently proposed dynamic model averaging (DMA) and dynamic model selection (DMS) framework to develop forecasting models of Chinese soybean futures price with eight predictors, which allows both coefficients and forecasting models to evolve over time. Specifically, covering an out-of-sample period from August 2, 2005 to May 26, 2017, experimental results show that the DMA and DMS outperform the time-varying parameter model, autoregressive model, linear regression (including all predictors), and random walk on the basis of the standard accuracy measures and Diebold-Mariano (DM) test. The best predictors for forecasting soybean futures price tend to be time-varying. Policymakers and investors should realize that there are many potential predictors whose predictive powers are strong but vary over time in Chinese soybean futures price forecasting.

KEY WORDS:

• Dynamic model averaging
• forecasting
• soybean futures
• time-varying

Supplementary material

Supplemental data for this article can be accessed here.

Notes

1. Statistical information are available at General Administration of Customs, China (http://www.customs.gov.cn/).

2. 1 CNY equals approximately 0.1478 USD.

3. Statistical information are available at Dalian Commodity Exchange (http://www.dce.com.cn/).

4. The optimal model is the model that has the highest forecasting performance in the past measured by the predictive density $p_k\left(\left.y_{t-1}\right|y_{t-2},y_{t-3},...,y_1\right)$.

5. The soybean net import is the difference between the soybean import and export (net import = import - export).

Additional information

Funding

This work was supported by the National Natural Science Foundation of China [71771101,71501079,71601147].