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 .
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.
The daily excess returns over the period (t, t+1),
it, from this trading strategy are
 obtained as follows:


where zt= +1 for long (buy signal) FC position
and zt = -1 for short (sell signal) FC
RESULTS –GBP /USD ( 1991 – 2000)
Measure
With transaction cost
Without
Transaction Cost
1 bp
2 bp
3 bp
Daily return
0.1627%
0.1527%
0.1427%
0.1327%
Annualized return
41.0110%
38.4910%
35.9710%
33.4510%
Annualized vol
21.9895%
21.9895%
21.9895%
21.9895%
792.3320913
743.6456913
694.9592913
646.2728913
1.865028187871280
1.750427871462690
1.635827555054100
1.521227238645500
Maximum daily profit
0.053053754
0.052953754
0.052853754
0.052753754
Maximum daily loss
-0.033799175
-0.033899175
-0.033999175
-0.034099175
53.36438923
53.05383023
52.95031056
52.69151139
46.63561077
46.94616977
47.04968944
47.30848861
cumulative return
Sharpe ratio
FORECASTING ACCURACY PERFORMANCE FOR
GBP /USD ( 1991 – 2000)
Model
RMSE
LS*
BVAR-TVP
0.029884
Random Walk
0.049023
MSE-T
-0.39649
ENC-T
20.90143
• 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.
27.9397
RESULTS –JPY/USD ( 1991 – 2000)
Measure
Without
Transaction Cost
With transaction cost
1 bp
2 bp
3 bp
Daily return
0.0611%
0.0511%
0.0411%
0.0311%
Annualized return
15.3903%
12.8703%
10.3503%
7.8303%
Annualized vol
24.4108%
24.4108%
24.4108%
24.4108%
285.644367
238.873167
192.101967
145.330767
0.630472317044453
0.527239240395165
0.424006163745878
0.320773087096594
Maximum daily profit
0.074769383
0.074669383
0.074569383
0.074469383
Maximum daily loss
-0.05107331
-0.05117331
-0.05127331
-0.05137331
51.83189655
51.67025862
51.45474138
51.34698276
48.16810345
48.32974138
48.54525862
48.65301724
cumulative return
Sharpe ratio
FORECASTING ACCURACY PERFORMANCE
JPY/USD ( 1991 – 2000)
Model
RMSE
LS*
BVAR-TVP
0.000232
Random Walk
0.037633
MSE-T
-0.79703
ENC-T
41.17992
• 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.
41.13595
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