1 financial informatics –ix: financial neural computing khurshid ahmad, professor of computer...
TRANSCRIPT
1
Financial Informatics –IX:
Financial Neural Computing
Khurshid Ahmad, Professor of Computer Science,
Department of Computer ScienceTrinity College,
Dublin-2, IRELANDNovember 17th, 2008.
https://www.cs.tcd.ie/Khurshid.Ahmad/Teaching.html
2
Financial Informatics: Neural Computing and
Volatility
Hamida, Shaikh A, and Zahid Iqbal (2004). Using neural networks for forecasting volatility of S&P 500 Index futures prices. Journal of Business Research Vol. 57, pp 1116– 1125
Financial time series models are used extensively in econometrics and in finance. However, pre-condition of the use of these models includes:
(a) Model Identification: whether to choose autoregressive or moving average or a hybrid of the two;
(b)The order of the model (c) The time series has minimal or noise.
3
Financial Informatics: Neural Computing and
Volatility
Hamida, Shaikh A, and Zahid Iqbal (2004). Using neural networks for forecasting volatility of S&P 500 Index futures prices. Journal of Business Research Vol. 57, pp 1116– 1125
Network Architecture:The number of neurons in the first layer—13—is equal tothe number of explanatory variables. We specified two timesthat many neurons in the second layer.
4
Financial Informatics: Neural Computing and
Volatility
Hamida, Shaikh A, and Zahid Iqbal (2004). Using neural networks for forecasting volatility of S&P 500 Index futures prices. Journal of Business Research Vol. 57, pp 1116– 1125
Training and Testing Data:We want to predict the volatility of the S&P 500 Indexfutures prices. Our raw data series consists of closing settlementprices of 16 nearest futures contracts and 3 spot indexes.We take the futures contract series that will mature in thenearest maturity month. The maturity months are March,June, September, and December#Future Contracts. Seven of the 16 futures
contracts are on commodities (silver, platinum, palladium,heating oil, copper, gold, crude oil), 3 are on Treasuryobligations (Treasury notes, bonds, and bills), and 6 are onforeign currencies (Swiss frank, yen, mark, Canadian dollar,British pound, euro dollar). The three spot indexes are DJIA,NYSE Composite Index, and S&P 500 Index.We also use 1-day lagged S&P 500 futures prices as an additional explanatoryvariable for a total of 20 variables. We select thesevariables because of availability of 10 years of daily data onthem—from February 1, 1984, to January 31, 1994—2531observations per variable. The data set was obtained fromKnight–Ridder Financial Publishing. Since neural networksneed to be trained with a large data set, it fits well with ourneeds. From the raw data series, we calculate 20-day rollinghistorical standard deviations (HSDs). We calculate HSDsfrom daily percentage price changes of the 20 variablescalculated as natural log relatives of the price or index series.The percentage change for Day 2 based on prices P1 and P2will be given by: ln( P2/P1).We use 500 HSD observations totrain the network and the rest for forecasting.
5
Financial Informatics: Neural Computing and
Volatility
Hamida, Shaikh A, and Zahid Iqbal (2004). Using neural networks for forecasting volatility of S&P 500 Index futures prices. Journal of Business Research Vol. 57, pp 1116– 1125
Future Contracts Elaboration Number
Commodities Silver, platinum, palladium, heating oil, copper, gold, crude oil
7
Treasury obligations Treasury notes, bonds, and bills 3
Foreign Currencies Swiss frank, yen, mark, Canadian dollar, British pound, euro dollar
6
Spot Indexes DJIA, NYSE Composite Index, and S&P 500 Index
3
Future Prices 1-Day lagged S&P futures 1
2531 data points were extracted from the 20 time series
6
Financial Informatics: Neural Computing and
Volatility
Hamida, Shaikh A, and Zahid Iqbal (2004). Using neural networks for forecasting volatility of S&P 500 Index futures prices. Journal of Business Research Vol. 57, pp 1116– 1125
Future Contracts
Elaboration Number
Commodities Silver, platinum, palladium, heating oil, copper, gold, crude oil
7
Treasury obligations
Treasury notes, bonds, and bills 3
Foreign Currencies
Swiss frank, yen, mark, Canadian dollar, British pound, euro dollar
6
Spot Indexes DJIA, NYSE Composite Index, and S&P 500 Index 3
Future Prices 1-Day lagged S&P futures 1
From the raw data series, we calculate 20-day rolling historical standard deviations (HSDs). We calculate HSDs from daily percentage price changes of the 20 variables calculated as natural log relatives of the price or index series. The percentage change for Day 2 based on prices P1 and P2 will be given by: ln( P2/P1).We use 500 HSD observations to train the network and the rest for forecasting.