Investigating the information content of non-cash-trading index futures using neural networks

Abstract

This study investigates the information content of SGX-DT Nikkei 225 and MSCI Taiwan index futures prices during the non-cash-trading (NCT) period. The lead–lag relationship between the futures market during the NCT period and the cash market during its opening period is first investigated by the generalized methods of moments. The obtained leading futures and previous day's cash market closing index are then used as the input variables to predict the opening cash price index by the backpropagation neural network model. Sensitivity analysis is first employed to address and solve the issue of finding the appropriate setup of the topology of the networks. Extensive studies are then performed on the robustness of the constructed network by using different training and testing sample sizes. To demonstrate the effectiveness of our proposed method, the 5-min intraday data of spot and futures index from a 6-month historical record were evaluated using the designed neural network model. Analytic results demonstrate that the proposed neural network model outperforms the neural network model with previous day's closing index as the input variable, the random walk and GARCH model forecasts. It, therefore, indicates that there is valuable information involved in the futures prices during the NCT period that can be used to forecast the opening cash price index. Besides, the neural network model provides better forecasting results than the commonly discussed GARCH model.

Introduction

It is quite common that the daily trading time of an index futures contract begins earlier and ends later than which of its underlying spot market. In this study, the time period between daily close and the subsequent day's opening of the cash index trading is defined as the non-cash-trading (NCT) period. Valuable information should be obtained through the analysis of the NCT futures and contribute to the success of exercising proper investment decisions for the underlying spot market. This research investigates the information content of SGX-DT (Singapore Exchange-Derivatives Trading Limited) Nikkei 225 and MSCI Taiwan index futures prices during the NCT period. Unlike past studies (Chan, 1992, Hiraki et al., 1995, Iihara et al., 1996, Martikainen and Puttonen, 1994, Min and Najand, 1999, Pizzi et al., 1998, Stoll and Whaley, 1990) only reported that index futures price changes lead price changes of the underlying spot market (the so-called lead–lag relationship between futures and cash market), this study tries to analyze this phenomena using a two-stage approach. It first tests if there is a lead–lag relationship between the index futures during the NCT period and the cash market index during its opening period. The obtained leading futures and previous day's cash market closing index are then used to predict the opening cash market price by the artificial neural networks (ANNs) model. The rationale underlying the analyses is to learn the information content of the NCT SGX-DT Nikkei 225 and MSCI Taiwan futures prices by comparing the opening cash price index forecast including NCT futures and the random walk¹ (with previous day's cash closing index as the forecasted opening cash price index) assumption. If the former forecasts outperform the latter then the information of the NCT index futures is considered valuable. Note that it is valuable to use the lead–lag relationship analysis as a supporting tool for neural networks as we can learn more about the inner workings. Besides, as there is no theoretical method in determining the best input variables, this procedure can be implemented as a generally accepted method for determining appropriate leading futures and thus giving statistical support in deciding the input vector of the designed neural network model.

Please note that we cannot use the popular ARIMA forecasting technique here since ARIMA is an univariate model which only uses the historical data to make inferences about the variable we are interested. And in this study we will be using both the cash index and futures index in predicting the opening cash price index and hence make the ARIMA approach inapplicable here. The ANNs are adopted in building the forecasting model with its ability to capture subtle functional relationships among the empirical data even though the underlying relationships are unknown or hard to describe. Besides, no strong model assumptions (variation homogeneity and system stationarity) are required and the literature on applying ANNs in the finance area is vast and fruitful (Vellido et al., 1999, Zhang et al., 1998). In this study, the backpropagation neural networks (BPN) with various numbers of nodes in the hidden layer and different learning rates is extensively studied to address and solve the issue of finding the appropriate setup of the topology of the networks. Further studies are performed on the robustness of the constructed networks using different training and testing sample sizes. To evaluate the effectiveness of the proposed neural network model, the daily 5-min transaction data of index futures and cash prices from October 1998 to March 1999 of SGX-DT Nikkei 225 and MSCI Taiwan index futures are used as illustrative examples. Finally, analytic results compared with BPN with previous day's closing index as the input, the random walk (with previous day's cash closing index as the forecasted opening cash price index), and GARCH model forecasts are also discussed.

The rest of the paper is organized as follows. We will give a brief overview of lead–lag relationship and neural networks in Section 2. The hypotheses and assumptions for the proposed study are presented in Section 3. Section 4 describes the lead–lag relationship analysis and the development of neural networks forecasting model. To verify the robustness of the designed neural network model, the prediction efficiency is summarized using different training and testing sample sizes in Section 5. Section 6 addresses the conclusion and possible future research areas.