Synthetic History for Exchange Traded Funds
To make money in trading one ought to forecast the future price, but to do so accurately one must verify the predictions using the past data. A short trading history can present a problem. We showed both theoretically and experimentally that the history of some financial assets can be reconstructed quite accurately. We forecasted the past price movements of exchange traded funds (ETFs). The problem in practice is very acute as there are a number of very liquid ETFs that can be traded with minimum slippage but their available history is too short. In such situations systematic traders cannot test their trading models as the history length is insufficient. To forecast historical ETF prices we used stocks with a longer history available. In some cases we created multiple model instances with a variable number of stocks. As soon as the stock history became unavailable we selected a different model. We compared this and eight other methods using a set of US ETFs ranging from S&P 500 to uranium. The experimental study showed the expectation maximisation with covariance matrix normalization to be the best method for this task.
Keywordssynthetic history artificial history time series regression expectation maximisation imputation missing data
Unable to display preview. Download preview PDF.
- 2.Graham, K.: Imputing for missing survey responses. s.l. In: Proceedings of the Section on Survey Research Methods. American Statistical Association (1982)Google Scholar
- 4.Little, R.J.A., Rubin, D.B.: Statistical Analysis with Missing Data, pp. 3–18, 39–48, 127–139. John Wiley & Sons, Los Angeles (1987)Google Scholar
- 6.Firat, M., Dikbas, F., Koc, A.C., Güngör, M.: Estimation of Missing River Flows using Expectation Maximization Method. Balwois, Ohrid (2010)Google Scholar
- 10.Hastie, T., Tibshirani, R., Friedman, J.: The Elements of Statistical Learning: Data Missing, Mining, Inference, and Prediction. Springer, New York (2001)Google Scholar
- 12.Srikanthan, R.: A multisite daily rainfall data generation model for climate change conditions. In: 18th World IMACS / MODSIM Congress, pp. 3976–3982. eWater CRC, Water Division, Bureau of Meteorology, Melbourne (2009)Google Scholar
- 14.Utsunomiya, K., Sonoda, K.: Methodology for Handling Missing Values In Tankan. Research and Statistics Department Bank of Japan, Japan (2001)Google Scholar
- 15.Bang, Y.-K., Lee, C.-H.: Fuzzy Time Series Prediction with Data Preprocessing and Error Compensation Based on Correlation Analysis. In: Third International Conference on Convergence and Hybrid Information Technology, vol. 2, pp. 714–721. IEEE (2008)Google Scholar
- 16.Shrestha, S.L.: Categorical Regression Models with Optimal Scaling for Predicting Indoor Air Pollution Concentrations inside Kitchens in Nepalese Households. Nepal Journal of Science and Technology 10, 205–211 (2009)Google Scholar
- 17.Sujatha, K.V., Sundaram, S.M.: Stock Index Prediction Using Regression and Neural Network Models under Non Normal Conditions. In: 2010 International Conference on Emerging Trends in Robotics and Communication Technologies (INTERACT), pp. 59–63 (2010)Google Scholar
- 19.Mustapha, N., Jalali, M., Bozorgniya, A., Jalali, M.: Navigation Patterns Mining Approach based on Expectation Maximization Algorithm. World Academy of Science, Engineering and Technology 50, 855–859 (2009)Google Scholar