09.03.2014 Views

FINANCIAL FORECASTING: Comparison Of ARIMA, FFNN And SVR

FINANCIAL FORECASTING: Comparison Of ARIMA, FFNN And SVR

FINANCIAL FORECASTING: Comparison Of ARIMA, FFNN And SVR

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

<strong>FINANCIAL</strong> <strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong> <strong>ARIMA</strong>, <strong>FFNN</strong> <strong>And</strong> <strong>SVR</strong><br />

Ashish Gajanan Lahane (05329R01)<br />

Under the guidance of<br />

Prof. Bernard L. Menezes<br />

July 8, 2008


Outline<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

1 Introduction<br />

2 Experiments <strong>And</strong> Results For <strong>ARIMA</strong><br />

3 Experiments <strong>And</strong> Results For <strong>FFNN</strong><br />

4 Experiments <strong>And</strong> Results For <strong>SVR</strong><br />

5 <strong>Comparison</strong><br />

6 Conclusion <strong>And</strong> Future Work<br />

7 APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

8 References


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

INTRODUCTION<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


Abbreviations<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Feed Forward Neural Networks<br />

AutoRegressive Integrated Moving Average<br />

Support Vector Regression<br />

<strong>SVR</strong> using Polynomial Kernel<br />

<strong>SVR</strong> using Radial Basis Function Kernel<br />

AutoCorrelation Function<br />

Partial AutoCorrelation function<br />

<strong>FFNN</strong><br />

<strong>ARIMA</strong><br />

<strong>SVR</strong><br />

<strong>SVR</strong> POLY<br />

<strong>SVR</strong> RBF<br />

ACF<br />

PACF


Literature Survey<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Financial time series: nonlinear, one of noisiest and most<br />

difficult signals to forecast<br />

For financial forecasting:<br />

<strong>FFNN</strong> better <strong>ARIMA</strong>[16, 7, 12, 13, 2, 11]<br />

<strong>SVR</strong> better than <strong>ARIMA</strong>[6]<br />

<strong>FFNN</strong> better than <strong>SVR</strong>[17, 6]<br />

<strong>SVR</strong> better than <strong>FFNN</strong>[4, 5]


Problem Definition<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

To compare three models:<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

for one-day-ahead forecasting performance on three important indices<br />

in Indian stock market:<br />

BSE Sensex<br />

CNX IT<br />

S&P CNX Nifty


Error Measures<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Mean Absolute Percent Error (MAPE)<br />

MAPE = 100 ×<br />

Mean Absolute Error (MAE)<br />

N∑<br />

|E t |<br />

t=1<br />

MAE =<br />

N<br />

where Y t ,E t represent desired outputs and corresponding errors<br />

at t=1,2,...,N respectively.<br />

Directional Symmetry (DS)<br />

DS = d correct<br />

N × 100<br />

where d correct = number of times the forecaster predicted the<br />

direction of the series right<br />

N∑<br />

t=1<br />

N<br />

|E t|<br />

Y t


Background<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

<strong>ARIMA</strong> go<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>FFNN</strong> go<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

<strong>SVR</strong> go


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

EXPERIMENTS AND<br />

RESULTS FOR <strong>ARIMA</strong><br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


ACF PACF with d = 0 For BSE Sensex<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


ACF PACF with d = 1 For BSE Sensex<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


ACF PACF with d = 2 For BSE Sensex<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


Model Identification<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

More than one differencing not needed<br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Summary<br />

Suitable models are<br />

AR(1) with φ 1 = 1 (i.e. <strong>ARIMA</strong>(0,1,0))<br />

AR(2)<br />

<strong>ARIMA</strong>(p, 1, q) with 0 ≤ p, q ≤ 2<br />

MA(q) (i.e. <strong>ARIMA</strong>(0,0,q)) models unsuitable<br />

0 ≤ p, d, q ≤ 2 should be tried


Results For <strong>ARIMA</strong><br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

BSE Sensex<br />

p d q MAPE DS<br />

1 1 1 1.351 79.49<br />

1 0 1 1.362 74.87<br />

1 1 0 1.406 77.95<br />

2 0 1 1.413 74.36<br />

2 0 0 1.424 77.95<br />

2 1 0 1.425 78.46<br />

1 0 0 1.428 76.41<br />

0 1 2 1.429 74.36<br />

1 0 2 1.430 76.41<br />

0 1 0 1.432 75.38<br />

0 1 1 1.435 76.41<br />

1 1 2 1.442 76.92<br />

2 1 1 1.468 74.87<br />

2 0 2 1.472 77.44<br />

2 1 2 1.661 74.36<br />

0 2 0 2.100 64.10<br />

1 2 0 2.754 65.13<br />

2 2 0 3.188 68.21<br />

2 2 2 20.078 58.46<br />

1 2 2 20.094 58.46<br />

0 2 2 21.819 55.38<br />

2 2 1 22.268 52.82<br />

1 2 1 22.776 55.38<br />

0 2 1 23.246 52.31<br />

0 0 2 24.551 50.26<br />

0 0 1 53.822 46.15<br />

0 0 0 65.824 41.03<br />

CNX IT<br />

p d q MAPE DS<br />

1 1 1 1.479 79.77<br />

1 0 1 1.480 79.10<br />

2 0 1 1.481 78.32<br />

1 0 0 1.482 77.00<br />

2 0 0 1.483 76.58<br />

2 1 0 1.500 75.89<br />

0 1 2 1.532 74.98<br />

1 1 2 1.544 76.85<br />

1 0 2 1.560 73.07<br />

0 1 0 1.585 75.00<br />

0 1 1 1.588 72.67<br />

1 1 0 1.598 72.83<br />

2 1 1 1.606 72.92<br />

2 0 2 1.612 71.11<br />

2 1 2 1.623 71.91<br />

0 2 0 2.166 71.12<br />

1 2 0 3.015 69.56<br />

2 2 0 3.861 68.77<br />

2 2 2 25.613 55.65<br />

1 2 2 42.084 50.62<br />

0 2 2 64.256 50.62<br />

2 2 1 77.953 50.62<br />

1 2 1 93.480 50.62<br />

0 2 1 93.493 50.62<br />

0 0 2 175.95 50.62<br />

0 0 1 183.54 50.62<br />

0 0 0 243.51 50.62<br />

S&P CNX Nifty<br />

p d q MAPE DS<br />

1 1 1 1.755 79.24<br />

2 0 2 1.813 77.08<br />

1 1 2 1.827 78.13<br />

1 1 0 1.833 79.17<br />

1 0 0 1.842 78.13<br />

2 0 0 1.856 77.08<br />

2 1 1 1.856 75.00<br />

2 1 2 1.862 77.08<br />

0 1 1 1.871 76.04<br />

0 1 2 1.873 75.00<br />

2 1 0 1.873 76.04<br />

1 0 2 1.888 77.08<br />

1 0 1 1.890 72.92<br />

0 1 0 1.901 77.08<br />

2 0 1 1.908 77.08<br />

0 2 0 2.477 71.88<br />

1 2 0 3.292 67.71<br />

2 2 0 3.695 66.67<br />

0 0 1 42.688 43.75<br />

1 2 2 43.182 52.08<br />

2 2 2 44.405 52.08<br />

0 2 1 45.573 52.08<br />

1 2 1 47.657 52.08<br />

2 2 1 53.251 52.08<br />

0 0 0 72.477 50.00<br />

0 0 2 78.234 47.42<br />

0 2 2 99.433 47.90<br />

Conclusion<br />

<strong>ARIMA</strong>(1,1,1) outperformed other <strong>ARIMA</strong> models


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

EXPERIMENTS AND<br />

RESULTS FOR <strong>FFNN</strong><br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>FFNN</strong> Parameters<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

<strong>FFNN</strong> Parameter<br />

Values Tried<br />

Hidden layers 1,2<br />

Trans func in hidden layer<br />

tanh, sigmoid<br />

Trans func in output layer<br />

tanh, sigmoid, linear<br />

[2 1],[3 1],[5 1],[7 1],[10 1],[15 1],[20 1],<br />

[40 1],[3 2 1],[5 2 1],[5 3 1],[7 2 1],<br />

Neurons per hidden layer [7 3 1],[7 5 1],[10 2 1],[10 3 1],<br />

[10 5 1],[10 7 1],[15 2 1],[15 3 1],<br />

[15 5 1],[15 7 1],[15 10 1]<br />

Window size 5,7,10,20,40,60,80<br />

Neurons IN output Layer 1<br />

6440 <strong>FFNN</strong> models per series


Results For <strong>FFNN</strong><br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Summary<br />

BSE<br />

S&P CNX<br />

Sensex CNX IT Nifty<br />

Min MAPE 1.1180 1.1863 1.0834<br />

Avg MAPE 13.0838 9.4293 10.8195<br />

BSE<br />

S&P CNX<br />

Sensex CNX IT Nifty<br />

Max DS 66.6667 62.1429 66.9524<br />

Avg DS 44.9146 47.8497 50.0907<br />

Very good performance for value forecasting<br />

Visible effect of random weight initialisation


Best <strong>FFNN</strong> Models<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Observation<br />

Number <strong>Of</strong><br />

Transfer<br />

Min Win Neurons Per function<br />

MAPE MAE DS Size Layer Per Layer<br />

BSE Sensex 1.1180 182.78 53.85 5 [7 2 1] [tanh tanh lin]<br />

CNX IT 1.1863 56.48 58.13 40 [10 1] [sigmoid lin]<br />

S&P CNX Nifty 1.0834 62.53 66.67 80 [15 2 1] [sigmoid sigmoid lin]<br />

Neurons Transfer<br />

Max Window Per function Per<br />

DS MAPE MAE Size Layer Layer<br />

BSE Sensex 66.6667 3.3060 606.3522 80 [20 1] [tanh tanh]<br />

CNX IT 62.1429 1.2795 61.0987 60 [3 1] [sigmoid lin]<br />

S&P CNX Nifty 66.9524 1.5371 88.4242 80 [2 1] [tanh lin]<br />

linear transfer function in last layer


Transfer Function In Last Layer<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

# of models with MAPE60<br />

and linear trans funct in<br />

Series # of models with DS>60 last layer %<br />

BSE Sensex 42 21 50.00<br />

CNX IT 9 7 77.78<br />

S&P CNX Nifty 134 74 55.22<br />

Conclusion<br />

linear transfer function in last layer gave good results


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

EXPERIMENTS AND<br />

RESULTS FOR <strong>SVR</strong><br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>SVR</strong> POLY Vs <strong>SVR</strong> RBF Min MAPE<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Avg MAPE<br />

<strong>SVR</strong> POLY <strong>SVR</strong> RBF <strong>SVR</strong> POLY <strong>SVR</strong> RBF<br />

BSE Sensex 1.4089 3.3324 2.5730 5.5208<br />

CNX IT 1.4695 3.8992 2.4327 6.9512<br />

S&P CNX Nifty 1.7230 1.7726 2.5094 3.5316<br />

Max DS<br />

Avg DS<br />

<strong>SVR</strong> POLY <strong>SVR</strong> RBF <strong>SVR</strong> POLY <strong>SVR</strong> RBF<br />

BSE Sensex 58.5492 50.1233 49.9497 40.2311<br />

CNX IT 55.2343 48.2390 49.9724 41.0412<br />

S&P CNX Nifty 61.9048 51.6624 54.1753 43.7640<br />

Conclusion<br />

<strong>SVR</strong> POLY outperforms <strong>SVR</strong> RBF


<strong>SVR</strong> POLY Parameter Analysis<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

BSE Sensex<br />

MAPE MAE DS C ε d γ r<br />

1.4089 234.1536 58.5492 249748.22 0.0213 2 1.2470 4.5355<br />

1.4880 257.8835 57.1429 792599.26 0.0106 2 2.5047 3.9518<br />

1.7269 294.6667 56.6667 402023.38 0.0184 2 2.2115 4.4671<br />

2.8427 499.2695 51.2500 296975.62 0.0119 3 0.5706 2.4815<br />

2.9433 489.9205 40.5128 936002.69 0.0099 2 2.8285 4.3638<br />

5.8682 992.6019 40.5263 575563.83 0.0075 4 3.7899 3.1718<br />

CNX IT<br />

MAPE MAE DS C ε d γ r<br />

1.4695 70.6894 55.2343 882549.15 0.0099 2 0.4234 1.9312<br />

1.4872 67.1975 52.4432 400000.00 0.0134 2 2.1210 3.3123<br />

1.5261 68.9893 54.9010 498213.28 0.0124 2 0.2328 2.7086<br />

3.7557 184.6057 43.4358 991231.61 0.0091 3 3.9946 2.4384<br />

3.8041 178.3055 44.1263 435563.83 0.0155 3 2.6472 4.2321<br />

S&P CXT Nifty<br />

MAPE MAE DS C ε d γ r<br />

1.7230 98.6894 61.9048 325315.10 0.0153 2 4.7430 0.9691<br />

1.7883 99.6453 57.1429 474668.02 0.0212 2 1.5805 1.4250<br />

2.2224 123.0545 58.3333 162817.77 0.0208 4 3.2448 3.2247<br />

2.4657 140.2473 45.6790 524259.16 0.0187 4 0.1475 3.9702<br />

3.5384 202.6957 57.4468 333194.98 0.0201 4 1.4261 1.3642<br />

3.6755 212.2441 56.0976 606613.45 0.0094 2 2.8840 4.4192<br />

Observations<br />

Best fit:<br />

polynomial<br />

kernel of<br />

degree<br />

d = 2<br />

Higher<br />

values of C<br />

Very small<br />

values of ε


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

COMPARISON<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>Comparison</strong> On Value Forecasting<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

MAPE <strong>ARIMA</strong> <strong>FFNN</strong> <strong>SVR</strong><br />

BSE Sensex 1.3513 1.1180 1.4089<br />

CNX IT 1.4788 1.1863 1.4695<br />

S&P CNX Nifty 1.7549 1.0834 1.7230<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Observation<br />

<strong>FFNN</strong> outperforms <strong>ARIMA</strong> and <strong>SVR</strong>


<strong>Comparison</strong> On Directional Forecasting<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

DS Naive Heuristic <strong>ARIMA</strong> <strong>FFNN</strong> SVM<br />

BSE Sensex 54.5455 79.4872 66.6667 58.5492<br />

CNX IT 52.5253 79.7718 62.1429 55.2343<br />

S&P CNX Nifty 51.0204 79.2417 66.9524 61.9048<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Observation<br />

<strong>ARIMA</strong> outperforms <strong>FFNN</strong> and <strong>SVR</strong>


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

CONCLUSION AND<br />

FUTURE WORK<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


Conclusion<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

<strong>ARIMA</strong>: <strong>ARIMA</strong>(1,1,1) is better fit than other <strong>ARIMA</strong> models<br />

<strong>FFNN</strong>: Models with linear transfer function in the last layer<br />

perform better<br />

<strong>SVR</strong>: <strong>SVR</strong> models with polynomial kernels of degree 2 are<br />

better fit than other <strong>SVR</strong> models<br />

Value Forecasting: <strong>FFNN</strong> models perform better than <strong>ARIMA</strong><br />

and <strong>SVR</strong> for these three series<br />

Directional Forecasting: <strong>ARIMA</strong> models perform better than<br />

<strong>FFNN</strong> and <strong>SVR</strong> for these three series


Future Work<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Addition of technical indicators such as Moving Average (MA),<br />

Moving Average Convergence/Divergence (MACD), On Balance<br />

Volume (OBV) etc[14, 18, 8]<br />

Addition of fundamental factors such as Price Earnings ratio<br />

(PE ratio), rate of change of company sales, Price Dividend<br />

ratio (PD ratio), rate of change of profit margin etc[1]<br />

N-days-ahead forecasting<br />

Forecasting on several other series<br />

Use of combining methods such as TopK, Best, DTopK, AFTER<br />

etc[19, 15]


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

THANKS!<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

APPENDIX<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

<strong>ARIMA</strong><br />

[appendix]<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>ARIMA</strong> Equation<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

<strong>ARIMA</strong>(p, d, q) model is given by:<br />

ϕ∇ d X t = θε t<br />

p∑<br />

where:- AR part: ϕ = 1 − ϕ i L i ,<br />

MA part: θ = 1 +<br />

i=1<br />

q∑<br />

θ j L j<br />

j=1<br />

I(difference) part: ∇ d = (1 − L 1 ) d<br />

Here X t−1 , X t−2 , ......X 2 , X 1 is the time series data<br />

L is lag operator, i.e. L i X t = X t−i ,<br />

ϕ i and θ j are the model parameters,<br />

ε t is a white noise process with zero mean and variance σ 2 .


<strong>ARIMA</strong> Methodology<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

1 Model Identification using ACF and PACF plots or other<br />

methods<br />

2 Model Estimation using methods such as likelihood or Bayesian<br />

3 Forecasting<br />

For further reading, please refer [3].<br />

go back


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

<strong>FFNN</strong><br />

[appendix]<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


A Simple Neuron<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

n∑<br />

y = φ( w i x i − θ)<br />

i=1


Transfer Functions<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

Linear<br />

Equation<br />

a = n<br />

range = range of<br />

n<br />

graphical<br />

representation:<br />

Sigmoid<br />

Equation<br />

a =<br />

1<br />

1 + e −n<br />

range = [0,+1]<br />

graphical<br />

representation:<br />

Tanh<br />

Equation<br />

a = tanh(n)<br />

range = [-1,+1]<br />

graphical<br />

representation:<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


<strong>FFNN</strong><br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

go back


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

<strong>SVR</strong><br />

[appendix]<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


Basic Principle<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Example 1: Classification<br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Example 2: Regression<br />

f (m 1 , m 2 , r) = C m 1m 2<br />

r 2<br />

log ⇓<br />

g(x, y, z) = ln(f (m 1 , m 2 , r)) = ln C + ln m 1 + ln m 2 − 2 ln r = c + x + y − 2z


SVM:Maximal Separating Hyperplane Classifier<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


SVM(1)<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Primal Form<br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Dual Form<br />

1<br />

minimize:<br />

2 ||w||2 ,<br />

subject to: c i (w · x i − b) ≥ 1, ∀ : 1 ≤ i ≤ n<br />

maximize:<br />

n∑<br />

α i − 1 ∑<br />

α i α j c i c j 〈x i · x j 〉<br />

2<br />

i=1<br />

i,j<br />

subject to: α i ≥ 0 and<br />

n∑<br />

α i c i = 0<br />

i=1


SVM(2)<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Solution To Dual Problem<br />

From the Solution we get<br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Soft Margin<br />

f =<br />

w = ∑ i<br />

α i c i x i<br />

n∑<br />

α i c i 〈x i · x j 〉 − b = ∑<br />

α i c i 〈x i · x j 〉 − b<br />

i=1<br />

i∈SV<br />

1<br />

minimize:<br />

2 ||w||2 + C ∑ ξ i ,<br />

i<br />

subject to: c i (w · x i − b) ≥ 1 − ξ i , ∀ : 1 ≤ i ≤ n.


<strong>SVR</strong><br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

subject to:<br />

minimize:<br />

⎧<br />

⎨<br />

⎩<br />

1<br />

2 ||w||2 + C<br />

l∑<br />

(ξ i + ξi ∗ )<br />

i=1<br />

y i − 〈w i , x i 〉 − b ≤ ε + ξ i<br />

〈w i , x i 〉 + b − y i ≤ ε + ξ ∗ i<br />

ξ i + ξ ∗ i ≥ 0


Kernels<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Radial Basis Function<br />

Equation<br />

k(x, x ′ ) = e (−γ|x−x′ | 2) ,<br />

for γ > 0<br />

graphical representation:<br />

Polynomial Function<br />

Equation<br />

k(x, x ′ ) = (γ(x · x ′ ) + r) d ,<br />

for γ > 0, r > 0<br />

graphical representation:<br />

go back


<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

REFERENCES<br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References


References I<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

A. Atiya, N. Talaat, and S. Shaheen.<br />

An efficient stock market forecasting model using neural networks.<br />

Neural Networks,1997., International Conference on, 4:2112–2115 vol.4, Jun 1997.<br />

E. Michael Azoff.<br />

Neural Network Time Series forecasting of Financial Markets.<br />

John Wiley & Sons, 1994.<br />

G. E. P. Box, G. M. Jenkins, and G. C. Reinsel.<br />

Time Series Analysis, Forecasting, and Control.<br />

Prentice-Hall, Englewood Cliffs, New Jersey, third edition, 1994.<br />

Lijuan Cao and Francis E.H Tay.<br />

Financial forecasting using support vector machines.<br />

Neural Computing & Applications, 10(2):184–192, May 2001.<br />

Chon Lung Chai.<br />

Finding kernel function for stock market prediction with support vector regression.<br />

Technical report, Universiti Teknologi Malaysia, 2006.<br />

Wun-Hua Chen, Jen-Ying Shih, and Soushan Wu.<br />

<strong>Comparison</strong> of support-vector machines and back propagation neural networks in forecasting the six major asian stock<br />

markets.<br />

International Journal of Electronic Finance, 1(1):49–67, January 2006.<br />

Wei Cheng, Lorry Wagner, and Chien-Hua Lin.<br />

Forecasting the 30-year u.s. treasury bond with a system of neural networks.


References II<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References<br />

Wei Cheng, Lorry Wagner, and Chien-Hua Lin.<br />

Forecasting the 30-year u.s. treasury bond with a system of neural networks.<br />

Shirley Gregor Feng Lin, Xing Huo Yu and Richard Irons.<br />

Time series forecasting with neural networks.<br />

Complexity International, 2, April 1995.<br />

Boyd M.S. Kermanshahi B. Kohzadi, N. and I. Kaastra.<br />

A comparison of artificial neural network and time series model for forecasting commodity price.<br />

Neurocomputing, 10:169–181, 1996.<br />

M. Kumar and M. Thenmozhi.<br />

Forecasting nifty index futures returns using neural network and arima models.<br />

Financial Engineering and Applications, 437, 2004.<br />

Jason E. Kutsurelis.<br />

Forecasting financial markets using neural networks: An analysis of methods and accuracy.<br />

Mohan Neeraj, Jha Pankaj, Laha Arnab Kumar, and Dutta Goutam.<br />

Artificial neural network models for forecasting stock price index in bombay stock exchange.<br />

IIMA Working Papers 2005-10-01, Indian Institute of Management Ahmedabad, Research and Publication Department,<br />

October 2005.<br />

available at http://ideas.repec.org/p/iim/iimawp/2005-10-01.html.<br />

Mohan Neeraj, Jha Pankaj, Laha Arnab Kumar, and Dutta Goutam.<br />

Artificial neural network models for forecasting stock price index in bombay stock exchange.<br />

IIMA Working Papers 2005-10-01, Indian Institute of Management Ahmedabad, Research and Publication Department,<br />

October 2005.


References III<br />

<strong>FINANCIAL</strong><br />

<strong>FORECASTING</strong>:<br />

<strong>Comparison</strong> <strong>Of</strong><br />

<strong>ARIMA</strong>, <strong>FFNN</strong><br />

<strong>And</strong> <strong>SVR</strong><br />

Introduction<br />

Experiments <strong>And</strong><br />

Results For<br />

<strong>ARIMA</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>FFNN</strong><br />

Experiments <strong>And</strong><br />

Results For <strong>SVR</strong><br />

<strong>Comparison</strong><br />

Conclusion <strong>And</strong><br />

Future Work<br />

Abhishek Seth.<br />

On using a multitude of time series forecasting models.<br />

Mtech thesis, Kanwal Rekhi School of Information Technology, IIT Bombay, 2006.<br />

Talaat N. Shaheen S. Atiya A.<br />

An efficient stock market forecasting model using neural networks.<br />

1997.<br />

Theodore B. Trafalis and Huseyin Ince.<br />

Support vector machine for regression and applications to financial forecasting.<br />

ijcnn, 06:6348, 2000.<br />

Chi-Cheong Chris Wong, Man-Chung Chan, and Chi-Chung Lam.<br />

Financial time series forecasting by neural network using conjugate gradient learning algorithm and multiple linear<br />

regression weight initialization.<br />

Computing in Economics and Finance 2000 61, Society for Computational Economics, July 2000.<br />

Y. Yang and H. Zou.<br />

Combining time series models for forecasting, 2002.<br />

APPENDIX<br />

<strong>ARIMA</strong><br />

<strong>FFNN</strong><br />

<strong>SVR</strong><br />

References

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!