Transcript Quantitative Trading Strategies
QUANTITATIVE TRADING STRATEGIES
ON THE SHORT-TERM PREDICTABILITY OF EXCHANGE RATES:A BVAR TIME-VARYING PARAMETERS APPROACH -NICHOLAS SARANTIS by Benziger Alice Priyanka Snehal Khair Prakash SuseendranVigeendharan Tiwari Ashutosh
PROCEDURES USED AND IMPLEMENTATION METHODOLOGIES APPLIED
implemented BVAR-TVP parameters in matlab Kalman implementation – Kalman toolbox in matlab Data – Bloomberg Optimization done for two parameters out of six (due to computation constraints), rest 4 parameters best fit value is used as per recommendation in paper
IMPROVISATIONS
The BVAR TVP parameters are regressed against recent data points ( last 1 month ) instead of the entire data points . Advantages Less Computations. Faster results.
More importance to recent Trends For GBP/USD This approach gives rise to higher annualized returns and less RMSE GBP/USD returns obtained are 41% and is better than the 5.7% returns obtained by using the approach mentioned in paper by author.
TRADING STRATEGY
The daily excess returns over the period (t, t+1), it, from this trading strategy are obtained as follows: where z t = +1 for long (buy signal) FC position and z t = -1 for short (sell signal) FC
RESULTS –GBP /USD ( 1991 – 2000)
Measure Without Transaction Cost With transaction cost Daily return Annualized return Annualized vol
0.1627%
41.0110%
21.9895%
1 bp
0.1527%
38.4910%
21.9895%
2 bp
0.1427%
35.9710%
21.9895%
3 bp
0.1327%
33.4510%
21.9895%
cumulative return Sharpe ratio Maximum daily profit
792.3320913
743.6456913
694.9592913
646.2728913
1.865028187871280
1.750427871462690
1.635827555054100
1.521227238645500
0.053053754
0.052953754
0.052853754
0.052753754
Maximum daily loss % winning trades % losing trades
-0.033799175
53.36438923
46.63561077
-0.033899175
53.05383023
46.94616977
-0.033999175
52.95031056
47.04968944
-0.034099175
52.69151139
47.30848861
FORECASTING ACCURACY PERFORMANCE FOR GBP /USD ( 1991 – 2000) RMSE Model BVAR-TVP Random Walk 0.029884
0.049023
LS* MSE-T ENC-T -0.39649
20.90143
27.9397
• RMSE obtained by BVAR-TVP model is less than random walk. Hence the prediction using this model is more accurate than a random walk model.
• RMSE Less than the RMSE obtained by the Author • Returns obtained by using the trading strategy mentioned earlier are substantial, suggesting model is accurate in prediction of FX rates.
RESULTS –JPY/USD ( 1991 – 2000)
Measure Without Transaction Cost With transaction cost Daily return Annualized return Annualized vol cumulative return Sharpe ratio Maximum daily profit Maximum daily loss % winning trades % losing trades
0.0611%
15.3903%
24.4108% 285.644367
0.630472317044453
0.074769383
-0.05107331
51.83189655
48.16810345
1 bp
0.0511%
12.8703%
24.4108% 238.873167
0.527239240395165
0.074669383
-0.05117331
51.67025862
48.32974138
2 bp
0.0411%
10.3503%
24.4108% 192.101967
0.424006163745878
0.074569383
-0.05127331
51.45474138
48.54525862
3 bp
0.0311%
7.8303%
24.4108% 145.330767
0.320773087096594
0.074469383
-0.05137331
51.34698276
48.65301724
FORECASTING ACCURACY PERFORMANCE JPY/USD ( 1991 – 2000) RMSE Model BVAR-TVP Random Walk 0.000232
0.037633
LS* MSE-T ENC-T -0.79703
41.17992
41.13595
• • RMSE obtained by BVAR-TVP model is less than random walk. Hence the prediction using this model is more accurate than a random walk model.
Returns obtained by using the strategy are low but substantial.
REFERENCES
Financial Econometrics Kalman Filter: some applications to Finance University of Evry - Master 2 Modelling and forecasting exchange rates with a Bayesian time-varying coefficient model Fabio Canova* http://www.cs.unc.edu/~welch/kalman/ http://www.cs.ubc.ca/~murphyk/Software/Kalma n/kalman_download.html
http://en.pudn.com/downloads158/sourcecode/o thers/detail706436_en.html