estimating and predicting stock returns using artificial neural networks dissertation paper...

31
Estimating and Estimating and Predicting Predicting Stock Returns Stock Returns Using Artificial Neural Using Artificial Neural Networks Networks Dissertation Paper Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE AND BANKING-DOFIN Supervisor: Prof. Univ. Dr. Moisa Altar MSc Student: Catalin-Marius Untea

Upload: egbert-clark

Post on 13-Dec-2015

222 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Estimating and Predicting Estimating and Predicting Stock Returns Stock Returns

Using Artificial Neural Using Artificial Neural NetworksNetworks

Dissertation PaperDissertation Paper

BUCHAREST ACADEMY OF ECONOMIC STUDIESDOCTORAL SCHOOL OF FINANCE AND BANKING-DOFIN

Supervisor:Prof. Univ. Dr. Moisa Altar

MSc Student:

Catalin-Marius Untea

Page 2: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

IntroductionIntroductionThis paper estimates and predicts stock returns, for shares traded on the Bucharest Stock Exchange, using both Artificial Neural Networks and Classical Econometric Arbitrage Pricing Theory (APT) methods, and compares the results obtained from both methods. The APT model is empirically implemented using Two-steep Cross Sectional Regression procedure introduced by Fama and McBeth (1973), and the One-step System of Non-linear Seemingly Unrelated Equations procedure firstly introduced by McElroy, Burmeister and Wall (1985). Neural network analyze, includes estimates conducted using Feedforward Neural Networks and Elman Recurrent Neural Networks.This paper does not try to discredit the Classical Econometric approach to the problem of estimation and prediction of stock returns, but it tries to emphasis the advantages brought by the new methods of estimation and prediction offered by Artificial Neural Networks, compared to classical econometrical methods with closed form used by many studies.What this paper is trying to bring additionally to other empirical studies in the field, is a complete practical approach to the problem of estimation and prediction, from the viewpoint of both econometric methods and neural network models. It concentrates on the shares traded on the Bucharest Stock Exchange, an emerging market during the last years.

Slide 2Slide 2

Page 3: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Theoretical background:Theoretical background: The Arbitrage Pricing Theory The Arbitrage Pricing Theory

(APT)(APT)

The Arbitrage Pricing Theory (APT) is a theoretical model, with tries to explain the behavior of stock returns to macroeconomic or firm specific factors.The major difficulty in applying the APT model comes from the fact that it shows that there is a method of predicting stock returns, but does not specify how exactly it must be solved. The main idea of the theory is that there exists a set of factors, so that, expected return can be expressed as a linear combination of those factors. The APT model is based on the hypotheses of arbitrage non-existence, which can be expressed as needing an upper limitation to the ratio between expected return and the volatility, of the same investment. If this ratio would not be limited, then it would be possible to obtain positive expected return for very low levels of risks.

Part IPart I

Slide 3Slide 3

Page 4: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

The APT model can be described by two equations:the first equation expresses the stock returns based on the set of factors

it

k

jjtijitit FbRER

1

][ , t=1,...,T i=1,...,N j=1,...,k

iR represents the return on share i;

][ iRE where represents the expected return for share i;

jF represents the influence of factor j on stock return i;

ijb

the second equation is for the equilibrium expected return and expresses the no arbitrage opportunity:

k

jjtijtit bRE

10][

where 0 represents the free-risk rate return;

j represents the risk premiums corresponding to risk factor j;

represents the sensitivity of the return on asset i to the fluctuations of factor j;

Theoretical background:Theoretical background: The Arbitrage Pricing Theory The Arbitrage Pricing Theory

(APT)(APT)Part IIPart II

Slide 4Slide 4

Page 5: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Two-step cross-sectional regression Two-step cross-sectional regression procedure procedure

Risk premium estimation for economic variables was introduced by Chen, Roll and Ross (1986), by making used of the two-step cross-sectional regression procedure, first introduced by Fama and McBeth. In the first stage of the procedure, the sensitivity coefficients for independent variables are estimated by making use of generalized method of moments (GMM).

During the first stage, the factor coefficients are estimated based on the following regression model:

itjtj

ijiit eFR

where itR represents portfolio return i;

jtF represents principal component j;

ij represents sensitivity coefficient for portfolio return i at factor fluctuations j.

Part IPart I

Slide 5Slide 5

Page 6: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT1Dependent Variable: RAND_PORT1

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:25Date: 06/24/07 Time: 16:25

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0103700.010370 0.003510.0035133

2.9516642.951664 0.00330.0033

COMP_PRIN1COMP_PRIN1 0.0010690.001069 0.003130.0031399

0.3406120.340612 0.73350.7335

COMP_PRIN2COMP_PRIN2 0.0006210.000621 0.004110.0041111

0.1511060.151106 0.87990.8799

COMP_PRIN3COMP_PRIN3 -0.000596-0.000596 0.002940.0029433

-0.202683-0.202683 0.83940.8394

COMP_PRIN4COMP_PRIN4 0.0018720.001872 0.005890.0058999

0.3173850.317385 0.75100.7510

COMP_PRIN5COMP_PRIN5 -0.003121-0.003121 0.004430.0044333

-0.703955-0.703955 0.48170.4817

R-squaredR-squared 0.0015080.001508 Mean dependent varMean dependent var 0.010340.0103400

Adjusted R-Adjusted R-squaredsquared

-0.005533-0.005533 S.D. dependent varS.D. dependent var 0.074860.0748644

S.E. of regressionS.E. of regression 0.0750710.075071 Sum squared residSum squared resid 3.995633.9956300

Durbin-Watson Durbin-Watson statstat

1.7945841.794584 J-statisticJ-statistic 7.31E-337.31E-33

Dependent Variable: RAND_PORT1Dependent Variable: RAND_PORT1

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:27Date: 06/24/07 Time: 16:27

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0002250.000225 0.0002220.000222 1.0121801.012180 0.31200.3120

COMP_PRIN1COMP_PRIN1 6.42E-066.42E-06 7.84E-067.84E-06 0.8193880.819388 0.41300.4130

COMP_PRIN2COMP_PRIN2 0.0001170.000117 0.0001190.000119 0.9878490.987849 0.32370.3237

COMP_PRIN3COMP_PRIN3 0.0001740.000174 0.0001760.000176 0.9891390.989139 0.32310.3231

COMP_PRIN4COMP_PRIN4 1.18E-061.18E-06 1.26E-051.26E-05 0.0936440.093644 0.92540.9254

COMP_PRIN5COMP_PRIN5 -2.48E-05-2.48E-05 2.74E-052.74E-05 -0.907004-0.907004 0.36490.3649

R-squaredR-squared 0.0024910.002491 Mean dependent varMean dependent var 0.000220.0002244

Adjusted R-Adjusted R-squaredsquared

-0.008143-0.008143 S.D. dependent varS.D. dependent var 0.004880.0048877

S.E. of regressionS.E. of regression 0.0049060.004906 Sum squared residSum squared resid 0.011290.0112900

Durbin-Watson Durbin-Watson statstat

2.0069552.006955 J-statisticJ-statistic 2.40E-282.40E-28

Estimation results for portfolio 1

2001 - 2003 2005 - 2006

Slide 6Slide 6

Page 7: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT2Dependent Variable: RAND_PORT2

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:22Date: 06/24/07 Time: 16:22

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0087330.008733 0.0030820.003082 2.8335672.833567 0.00470.0047

COMP_PRIN1COMP_PRIN1 -0.002338-0.002338 0.0039640.003964 -0.589960-0.589960 0.55540.5554

COMP_PRIN2COMP_PRIN2 -0.004659-0.004659 0.0029040.002904 -1.604640-1.604640 0.10900.1090

COMP_PRIN3COMP_PRIN3 -0.002547-0.002547 0.0028320.002832 -0.899318-0.899318 0.36880.3688

COMP_PRIN4COMP_PRIN4 -0.001160-0.001160 0.0044850.004485 -0.258574-0.258574 0.79600.7960

COMP_PRIN5COMP_PRIN5 -0.002081-0.002081 0.0024950.002495 -0.834141-0.834141 0.40450.4045

R-squaredR-squared 0.0117440.011744 Mean dependent varMean dependent var 0.008560.0085677

Adjusted R-Adjusted R-squaredsquared

0.0047740.004774 S.D. dependent varS.D. dependent var 0.056580.0565877

S.E. of regressionS.E. of regression 0.0564510.056451 Sum squared residSum squared resid 2.259402.2594044

Durbin-Watson Durbin-Watson statstat

1.4271651.427165 J-statisticJ-statistic 1.55E-311.55E-31

Dependent Variable: RAND_PORT2Dependent Variable: RAND_PORT2

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:23Date: 06/24/07 Time: 16:23

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC -2.12E-33-2.12E-33 6.68E-196.68E-19 -3.17E-15-3.17E-15 1.00001.0000

COMP_PRIN1COMP_PRIN1 3.08E-333.08E-33 6.03E-196.03E-19 5.11E-155.11E-15 1.00001.0000

COMP_PRIN2COMP_PRIN2 -1.00E-32-1.00E-32 1.62E-181.62E-18 -6.20E-15-6.20E-15 1.00001.0000

COMP_PRIN3COMP_PRIN3 4.62E-334.62E-33 1.02E-181.02E-18 4.53E-154.53E-15 1.00001.0000

COMP_PRIN4COMP_PRIN4 1.39E-321.39E-32 2.64E-182.64E-18 5.26E-155.26E-15 1.00001.0000

COMP_PRIN5COMP_PRIN5 1.16E-331.16E-33 1.79E-181.79E-18 6.45E-166.45E-16 1.00001.0000

Mean dependent Mean dependent varvar

0.0000000.000000 S.D. dependent varS.D. dependent var 0.000000.0000000

S.E. of regressionS.E. of regression 1.74E-321.74E-32 Sum squared residSum squared resid 1.41E-611.41E-61

Durbin-Watson Durbin-Watson statstat

1.7151211.715121 J-statisticJ-statistic 0.049420.0494200

Estimation results for portfolio 2

2001 - 2003 2005 - 2006

Slide 7Slide 7

Page 8: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT3Dependent Variable: RAND_PORT3

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:18Date: 06/24/07 Time: 16:18

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0067100.006710 0.0024600.002460 2.7279742.727974 0.00650.0065

COMP_PRIN1COMP_PRIN1 -0.000543-0.000543 0.0029280.002928 -0.185520-0.185520 0.85290.8529

COMP_PRIN2COMP_PRIN2 0.0033820.003382 0.0036590.003659 0.9244280.924428 0.35560.3556

COMP_PRIN3COMP_PRIN3 -0.004555-0.004555 0.0035140.003514 -1.296118-1.296118 0.19540.1954

COMP_PRIN4COMP_PRIN4 0.0020280.002028 0.0050670.005067 0.4002080.400208 0.68910.6891

COMP_PRIN5COMP_PRIN5 -0.002333-0.002333 0.0041690.004169 -0.559522-0.559522 0.57600.5760

R-squaredR-squared 0.0060250.006025 Mean dependent varMean dependent var 0.006930.0069377

Adjusted R-Adjusted R-squaredsquared

-0.000985-0.000985 S.D. dependent varS.D. dependent var 0.066360.0663688

S.E. of regressionS.E. of regression 0.0664010.066401 Sum squared residSum squared resid 3.126023.1260222

Durbin-Watson Durbin-Watson statstat

1.9757621.975762 J-statisticJ-statistic 1.10E-311.10E-31

Dependent Variable: RAND_PORT3Dependent Variable: RAND_PORT3

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:20Date: 06/24/07 Time: 16:20

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0028040.002804 0.0041180.004118 0.6808920.680892 0.49630.4963

COMP_PRIN1COMP_PRIN1 0.0016720.001672 0.0010250.001025 1.6314851.631485 0.10350.1035

COMP_PRIN2COMP_PRIN2 -0.011449-0.011449 0.0038520.003852 -2.971858-2.971858 0.00310.0031

COMP_PRIN3COMP_PRIN3 -0.008726-0.008726 0.0033180.003318 -2.629798-2.629798 0.00880.0088

COMP_PRIN4COMP_PRIN4 0.0054820.005482 0.0035420.003542 1.5476951.547695 0.12240.1224

COMP_PRIN5COMP_PRIN5 0.0061240.006124 0.0049810.004981 1.2294811.229481 0.21950.2195

R-squaredR-squared 0.0347140.034714 Mean dependent varMean dependent var 0.003430.0034311

Adjusted R-Adjusted R-squaredsquared

0.0244230.024423 S.D. dependent varS.D. dependent var 0.098170.0981788

S.E. of regressionS.E. of regression 0.0969720.096972 Sum squared residSum squared resid 4.410234.4102300

Durbin-Watson Durbin-Watson statstat

2.0668032.066803 J-statisticJ-statistic 5.30E-315.30E-31

Estimation results for portfolio 3

2001 - 2003 2005 - 2006

Slide 8Slide 8

Page 9: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT4Dependent Variable: RAND_PORT4

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:13Date: 06/24/07 Time: 16:13

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef Convergence achieved after: 1 weight matrix, 2 total coef iterationsiterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0138910.013891 0.004260.0042600

3.2612723.261272 0.00120.0012

COMP_PRIN1COMP_PRIN1 -4.93E-05-4.93E-05 0.004570.0045733

-0.010774-0.010774 0.99140.9914

COMP_PRIN2COMP_PRIN2 -0.005801-0.005801 0.004750.0047500

-1.221195-1.221195 0.22240.2224

COMP_PRIN3COMP_PRIN3 -0.017844-0.017844 0.003960.0039644

-4.501326-4.501326 0.00000.0000

COMP_PRIN4COMP_PRIN4 0.0003190.000319 0.005530.0055388

0.0576470.057647 0.95400.9540

COMP_PRIN5COMP_PRIN5 0.0057380.005738 0.006920.0069277

0.8283800.828380 0.40770.4077

R-squaredR-squared 0.0356630.035663 Mean dependent Mean dependent varvar

0.014010.0140111

Adjusted R-Adjusted R-squaredsquared

0.0288630.028863 S.D. dependent varS.D. dependent var 0.102160.1021600

S.E. of regressionS.E. of regression 0.1006750.100675 Sum squared residSum squared resid 7.186017.1860133

Durbin-Watson Durbin-Watson statstat

1.8835551.883555 J-statisticJ-statistic 4.66E-314.66E-31

Dependent Variable: RAND_PORT4Dependent Variable: RAND_PORT4

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:15Date: 06/24/07 Time: 16:15

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC -0.002295-0.002295 0.0047760.004776 -0.480618-0.480618 0.63100.6310

COMP_PRIN1COMP_PRIN1 -0.001392-0.001392 0.0015190.001519 -0.916460-0.916460 0.35990.3599

COMP_PRIN2COMP_PRIN2 -0.012073-0.012073 0.0039250.003925 -3.076084-3.076084 0.00220.0022

COMP_PRIN3COMP_PRIN3 -0.010983-0.010983 0.0039340.003934 -2.791577-2.791577 0.00550.0055

COMP_PRIN4COMP_PRIN4 0.0011900.001190 0.0046290.004629 0.2570300.257030 0.79730.7973

COMP_PRIN5COMP_PRIN5 0.0044510.004451 0.0025420.002542 1.7511891.751189 0.08060.0806

R-squaredR-squared 0.0351290.035129 Mean dependent varMean dependent var --0.00.001018787

33

Adjusted R-Adjusted R-squaredsquared

0.0248430.024843 S.D. dependent varS.D. dependent var 0.104250.1042599

S.E. of regressionS.E. of regression 0.1029560.102956 Sum squared residSum squared resid 4.971374.9713733

Durbin-Watson Durbin-Watson statstat

2.0685892.068589 J-statisticJ-statistic 7.70E-327.70E-32

Estimation results for portfolio 4

2001 - 2003 2005 - 2006

Slide 9Slide 9

Page 10: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT5Dependent Variable: RAND_PORT5

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:06Date: 06/24/07 Time: 16:06

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef Convergence achieved after: 1 weight matrix, 2 total coef iterationsiterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0116290.011629 0.004880.0048888

2.3793642.379364 0.01760.0176

COMP_PRIN1COMP_PRIN1 0.0062900.006290 0.005290.0052955

1.1880061.188006 0.23520.2352

COMP_PRIN2COMP_PRIN2 -0.010741-0.010741 0.005000.0050066

-2.145833-2.145833 0.03220.0322

COMP_PRIN3COMP_PRIN3 -0.012818-0.012818 0.004200.0042055

-3.048595-3.048595 0.00240.0024

COMP_PRIN4COMP_PRIN4 -0.003786-0.003786 0.006710.0067122

-0.564091-0.564091 0.57290.5729

COMP_PRIN5COMP_PRIN5 -0.001225-0.001225 0.005700.0057099

-0.214617-0.214617 0.83010.8301

R-squaredR-squared 0.0252550.025255 Mean dependent Mean dependent varvar

0.011390.0113900

Adjusted R-Adjusted R-squaredsquared

0.0183800.018380 S.D. dependent varS.D. dependent var 0.108750.1087599

S.E. of regressionS.E. of regression 0.1077540.107754 Sum squared residSum squared resid 8.232218.2322188

Durbin-Watson Durbin-Watson statstat

1.6886301.688630 J-statisticJ-statistic 4.69E-304.69E-30

Dependent Variable: RAND_PORT5Dependent Variable: RAND_PORT5

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:08Date: 06/24/07 Time: 16:08

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0048610.004861 0.0043640.004364 1.1140301.114030 0.26580.2658

COMP_PRIN1COMP_PRIN1 -0.002308-0.002308 0.0014530.001453 -1.588477-1.588477 0.11290.1129

COMP_PRIN2COMP_PRIN2 -0.022491-0.022491 0.0051280.005128 -4.386342-4.386342 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.040446-0.040446 0.0070500.007050 -5.737089-5.737089 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.000495-0.000495 0.0045230.004523 -0.109453-0.109453 0.91290.9129

COMP_PRIN5COMP_PRIN5 -0.001721-0.001721 0.0048940.004894 -0.351652-0.351652 0.72530.7253

R-squaredR-squared 0.1791570.179157 Mean dependent varMean dependent var 0.004630.0046300

Adjusted R-Adjusted R-squaredsquared

0.1704060.170406 S.D. dependent varS.D. dependent var 0.127860.1278677

S.E. of regressionS.E. of regression 0.1164640.116464 Sum squared residSum squared resid 6.361406.3614033

Durbin-Watson Durbin-Watson statstat

2.3237142.323714 J-statisticJ-statistic 3.19E-313.19E-31

Estimation results for portfolio 5

2001 - 2003 2005 - 2006

Slide 10Slide 10

Page 11: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT6Dependent Variable: RAND_PORT6

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:00Date: 06/24/07 Time: 16:00

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef Convergence achieved after: 1 weight matrix, 2 total coef iterationsiterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficientCoefficient Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0188450.018845 0.004700.0047099

4.0022434.002243 0.00010.0001

COMP_PRIN1COMP_PRIN1 0.0057470.005747 0.006620.0066211

0.8679550.867955 0.38570.3857

COMP_PRIN2COMP_PRIN2 -0.013967-0.013967 0.006300.0063099

-2.213911-2.213911 0.02720.0272

COMP_PRIN3COMP_PRIN3 -0.028096-0.028096 0.006490.0064922

-4.327917-4.327917 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.009366-0.009366 0.008080.0080811

-1.159018-1.159018 0.24680.2468

COMP_PRIN5COMP_PRIN5 0.0046710.004671 0.008350.0083555

0.5590610.559061 0.57630.5763

R-squaredR-squared 0.0709400.070940 Mean dependent Mean dependent varvar

0.018970.0189744

Adjusted R-Adjusted R-squaredsquared

0.0643880.064388 S.D. dependent varS.D. dependent var 0.123350.1233544

S.E. of regressionS.E. of regression 0.1193170.119317 Sum squared residSum squared resid 10.093710.093722

Durbin-Watson Durbin-Watson statstat

1.8306131.830613 J-statisticJ-statistic 7.01E-317.01E-31

Dependent Variable: RAND_PORT6Dependent Variable: RAND_PORT6

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 16:04Date: 06/24/07 Time: 16:04

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficientCoefficient Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0095140.009514 0.005310.0053111

1.7913331.791333 0.07390.0739

COMP_PRIN1COMP_PRIN1 0.0013400.001340 0.001360.0013699

0.9782500.978250 0.32850.3285

COMP_PRIN2COMP_PRIN2 -0.030951-0.030951 0.004950.0049500

-6.253226-6.253226 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.061180-0.061180 0.006440.0064433

-9.495897-9.495897 0.00000.0000

COMP_PRIN4COMP_PRIN4 0.0041930.004193 0.004270.0042799

0.9798560.979856 0.32770.3277

COMP_PRIN5COMP_PRIN5 0.0066950.006695 0.003990.0039977

1.6751081.675108 0.09460.0946

R-squaredR-squared 0.2955580.295558 Mean dependent Mean dependent varvar

0.009080.0090866

Adjusted R-Adjusted R-squaredsquared

0.2880480.288048 S.D. dependent varS.D. dependent var 0.146430.1464388

S.E. of regressionS.E. of regression 0.1235600.123560 Sum squared residSum squared resid 7.160257.1602566

Durbin-Watson Durbin-Watson statstat

2.0951712.095171 J-statisticJ-statistic 2.32E-322.32E-32

Estimation results for portfolio 6

2001 - 2003 2005 - 2006

Slide 11Slide 11

Page 12: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT7Dependent Variable: RAND_PORT7

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:50Date: 06/24/07 Time: 15:50

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0159130.015913 0.0042370.004237 3.7559093.755909 0.00020.0002

COMP_PRIN1COMP_PRIN1 -0.001386-0.001386 0.0055200.005520 -0.251057-0.251057 0.80180.8018

COMP_PRIN2COMP_PRIN2 -0.013094-0.013094 0.0047410.004741 -2.761637-2.761637 0.00590.0059

COMP_PRIN3COMP_PRIN3 -0.025616-0.025616 0.0056110.005611 -4.565651-4.565651 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.002776-0.002776 0.0061310.006131 -0.452871-0.452871 0.65080.6508

COMP_PRIN5COMP_PRIN5 -0.009707-0.009707 0.0061740.006174 -1.572245-1.572245 0.11630.1163

R-squaredR-squared 0.0730870.073087 Mean dependent varMean dependent var 0.015800.0158066

Adjusted R-Adjusted R-squaredsquared

0.0665500.066550 S.D. dependent varS.D. dependent var 0.109230.1092344

S.E. of regressionS.E. of regression 0.1055370.105537 Sum squared residSum squared resid 7.896857.8968511

Durbin-Watson Durbin-Watson statstat

1.8450981.845098 J-statisticJ-statistic 2.88E-302.88E-30

Dependent Variable: RAND_PORT7Dependent Variable: RAND_PORT7

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:52Date: 06/24/07 Time: 15:52

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0096330.009633 0.0041510.004151 2.3204282.320428 0.02070.0207

COMP_PRIN1COMP_PRIN1 -0.000167-0.000167 0.0010060.001006 -0.165679-0.165679 0.86850.8685

COMP_PRIN2COMP_PRIN2 -0.023277-0.023277 0.0044320.004432 -5.252230-5.252230 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.038118-0.038118 0.0053670.005367 -7.102547-7.102547 0.00000.0000

COMP_PRIN4COMP_PRIN4 0.0016510.001651 0.0032840.003284 0.5027870.502787 0.61530.6153

COMP_PRIN5COMP_PRIN5 -3.72E-05-3.72E-05 0.0036820.003682 -0.010109-0.010109 0.99190.9919

R-squaredR-squared 0.2383830.238383 Mean dependent varMean dependent var 0.009630.0096300

Adjusted R-Adjusted R-squaredsquared

0.2302640.230264 S.D. dependent varS.D. dependent var 0.106020.1060299

S.E. of regressionS.E. of regression 0.0930240.093024 Sum squared residSum squared resid 4.058464.0584600

Durbin-Watson Durbin-Watson statstat

2.0966402.096640 J-statisticJ-statistic 3.21E-313.21E-31

Estimation results for portfolio 7

2001 - 2003 2005 - 2006

Slide 12Slide 12

Page 13: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT8Dependent Variable: RAND_PORT8

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:19Date: 06/24/07 Time: 15:19

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef Convergence achieved after: 1 weight matrix, 2 total coef iterationsiterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0222620.022262 0.005420.0054299

4.1004484.100448 0.00000.0000

COMP_PRIN1COMP_PRIN1 -0.001124-0.001124 0.008180.0081800

-0.137423-0.137423 0.89070.8907

COMP_PRIN2COMP_PRIN2 -0.028633-0.028633 0.007730.0077322

-3.703067-3.703067 0.00020.0002

COMP_PRIN3COMP_PRIN3 -0.061396-0.061396 0.007950.0079522

-7.720936-7.720936 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.012017-0.012017 0.010590.0105988

-1.133904-1.133904 0.25720.2572

COMP_PRIN5COMP_PRIN5 0.0015340.001534 0.008000.0080022

0.1917270.191727 0.84800.8480

R-squaredR-squared 0.2076510.207651 Mean dependent Mean dependent varvar

0.022440.0224477

Adjusted R-Adjusted R-squaredsquared

0.2020630.202063 S.D. dependent varS.D. dependent var 0.153200.1532077

S.E. of regressionS.E. of regression 0.1368560.136856 Sum squared residSum squared resid 13.279213.279211

Durbin-Watson Durbin-Watson statstat

1.8644001.864400 J-statisticJ-statistic 3.95E-323.95E-32

Dependent Variable: RAND_PORT8Dependent Variable: RAND_PORT8

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:20Date: 06/24/07 Time: 15:20

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0216350.021635 0.0053450.005345 4.0479774.047977 0.00010.0001

COMP_PRIN1COMP_PRIN1 -0.000399-0.000399 0.0012270.001227 -0.324855-0.324855 0.74540.7454

COMP_PRIN2COMP_PRIN2 -0.045252-0.045252 0.0048960.004896 -9.242509-9.242509 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.089708-0.089708 0.0073880.007388 -12.14312-12.14312 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.004756-0.004756 0.0038320.003832 -1.241197-1.241197 0.21520.2152

COMP_PRIN5COMP_PRIN5 0.0094630.009463 0.0039240.003924 2.4116392.411639 0.01630.0163

R-squaredR-squared 0.5065220.506522 Mean dependent varMean dependent var 0.020560.0205677

Adjusted R-Adjusted R-squaredsquared

0.5012610.501261 S.D. dependent varS.D. dependent var 0.163420.1634233

S.E. of regressionS.E. of regression 0.1154120.115412 Sum squared residSum squared resid 6.247046.2470433

Durbin-Watson Durbin-Watson statstat

2.0232082.023208 J-statisticJ-statistic 2.08E-312.08E-31

Estimation results for portfolio 8

2001 - 2003 2005 - 2006

Slide 13Slide 13

Page 14: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: RAND_PORT9Dependent Variable: RAND_PORT9

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:18Date: 06/24/07 Time: 15:18

Sample: 252 966Sample: 252 966

Included observations: 715Included observations: 715

Kernel: Bartlett, Bandwidth: Fixed (6), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (6), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0126720.012672 0.0043240.004324 2.9307342.930734 0.00350.0035

COMP_PRIN1COMP_PRIN1 0.0090910.009091 0.0061600.006160 1.4759421.475942 0.14040.1404

COMP_PRIN2COMP_PRIN2 -0.028994-0.028994 0.0055710.005571 -5.204247-5.204247 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.055857-0.055857 0.0065060.006506 -8.585827-8.585827 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.010915-0.010915 0.0070770.007077 -1.542251-1.542251 0.12350.1235

COMP_PRIN5COMP_PRIN5 -0.010612-0.010612 0.0067690.006769 -1.567750-1.567750 0.11740.1174

R-squaredR-squared 0.2656820.265682 Mean dependent varMean dependent var 0.012540.0125400

Adjusted R-Adjusted R-squaredsquared

0.2605030.260503 S.D. dependent varS.D. dependent var 0.125230.1252311

S.E. of regressionS.E. of regression 0.1076910.107691 Sum squared residSum squared resid 8.222598.2225944

Durbin-Watson Durbin-Watson statstat

1.8746111.874611 J-statisticJ-statistic 2.93E-312.93E-31

Dependent Variable: RAND_PORT9Dependent Variable: RAND_PORT9

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 15:16Date: 06/24/07 Time: 15:16

Sample: 1240 1714Sample: 1240 1714

Included observations: 475Included observations: 475

Kernel: Bartlett, Bandwidth: Fixed (5), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (5), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3Instrument list: COMP_PRIN1 COMP_PRIN2 COMP_PRIN3

COMP_PRIN4 COMP_PRIN5COMP_PRIN4 COMP_PRIN5

VariableVariable CoefficienCoefficientt

Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0126200.012620 0.0044260.004426 2.8515112.851511 0.00450.0045

COMP_PRIN1COMP_PRIN1 -0.006055-0.006055 0.0059330.005933 -1.020475-1.020475 0.30800.3080

COMP_PRIN2COMP_PRIN2 -0.057778-0.057778 0.0083770.008377 -6.897398-6.897398 0.00000.0000

COMP_PRIN3COMP_PRIN3 -0.090673-0.090673 0.0056440.005644 -16.06432-16.06432 0.00000.0000

COMP_PRIN4COMP_PRIN4 -0.018203-0.018203 0.0155710.015571 -1.169005-1.169005 0.24300.2430

COMP_PRIN5COMP_PRIN5 0.0245970.024597 0.0165010.016501 1.4906651.490665 0.13670.1367

R-squaredR-squared 0.5902840.590284 Mean dependent varMean dependent var 0.011880.0118811

Adjusted R-Adjusted R-squaredsquared

0.5859160.585916 S.D. dependent varS.D. dependent var 0.168000.1680033

S.E. of regressionS.E. of regression 0.1081090.108109 Sum squared residSum squared resid 5.481455.4814533

Durbin-Watson Durbin-Watson statstat

2.0891962.089196 J-statisticJ-statistic 6.06E-316.06E-31

Estimation results for portfolio 9

2001 - 2003 2005 - 2006

Slide 14Slide 14

Page 15: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Two-step cross-sectional Two-step cross-sectional regression procedure regression procedure

Part IIPart II

At the second stage, the estimated sensitivity coefficients are used as independent variables in the cross-sectional regression in order to estimate the risk premium of the observed variables.The previously estimated sensitivity coefficients β, in the first stage, are used in the cross-section regression as independent variables, and portfolios mean returns are used as dependent variables. Each coefficient obtained by estimating the cross-section regression, represents an estimation for the risk premium associated to the exposure to unexpected variation in one of the factors.

u

returnportfoliomean

PRINCOMPPRINCOMP

PRINCOMPPRINCOMPPRINCOMPPRINCOMPPRINCOMPPRINCOMP

PRINCOMPPRINCOMP

5_5_

4_4_3_3_2_2_

1_1_1

ˆˆ

ˆˆˆˆˆˆ

ˆˆˆ__

Slide 15Slide 15

Page 16: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Dependent Variable: MEDII_PORT_01_03Dependent Variable: MEDII_PORT_01_03

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 18:17Date: 06/24/07 Time: 18:17

Sample: 1 9Sample: 1 9

Included observations: 9Included observations: 9

Kernel: Bartlett, Bandwidth: Fixed (2), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (2), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef iterationsConvergence achieved after: 1 weight matrix, 2 total coef iterations

Instrument list: BETA_COMP_PRIN1_01_03 BETA_COMP_PRIN2_01_Instrument list: BETA_COMP_PRIN1_01_03 BETA_COMP_PRIN2_01_

03 BETA_COMP_PRIN3_01_03 BETA_COMP_PRIN4_01_0303 BETA_COMP_PRIN3_01_03 BETA_COMP_PRIN4_01_03

BETA_COMP_PRIN5_01_03BETA_COMP_PRIN5_01_03

VariableVariable CoefficientCoefficient Std. ErrorStd. Error t-Statistict-Statistic Prob. Prob.

CC 0.0105910.010591 0.0017420.001742 6.0811216.081121 0.00890.0089

BETA_COMP_PRINBETA_COMP_PRIN1_01_031_01_03

-0.359591-0.359591 0.2491040.249104 --1.41.4

4354353737

0.24460.2446

BETA_COMP_PRINBETA_COMP_PRIN2_01_032_01_03

-0.086729-0.086729 0.2276780.227678 --0.30.3

8098092626

0.72860.7286

BETA_COMP_PRINBETA_COMP_PRIN3_01_033_01_03

-0.064155-0.064155 0.0983610.098361 --0.60.6

5225223737

0.56080.5608

BETA_COMP_PRINBETA_COMP_PRIN4_01_034_01_03

-0.414972-0.414972 0.3956120.395612 --1.01.0

4894893636

0.37130.3713

BETA_COMP_PRINBETA_COMP_PRIN5_01_035_01_03

0.3152640.315264 0.1374190.137419 2.2941872.294187 0.10550.1055

R-squaredR-squared 0.7832100.783210 Mean dependent Mean dependent varvar

0.013440.0134466

Adjusted R-Adjusted R-squaredsquared

0.4218930.421893 S.D. dependent varS.D. dependent var 0.004970.0049799

S.E. of regressionS.E. of regression 0.0037860.003786 Sum squared residSum squared resid 4.30E-054.30E-05

Durbin-Watson Durbin-Watson statstat

1.4223911.422391 J-statisticJ-statistic 6.18E-296.18E-29

Dependent Variable: MEDII_PORT_05_06Dependent Variable: MEDII_PORT_05_06

Method: Generalized Method of MomentsMethod: Generalized Method of Moments

Date: 06/24/07 Time: 18:51Date: 06/24/07 Time: 18:51

Sample: 1 9Sample: 1 9

Included observations: 9Included observations: 9

Kernel: Bartlett, Bandwidth: Fixed (2), No prewhiteningKernel: Bartlett, Bandwidth: Fixed (2), No prewhitening

Simultaneous weighting matrix & coefficient iterationSimultaneous weighting matrix & coefficient iteration

Convergence achieved after: 1 weight matrix, 2 total coef Convergence achieved after: 1 weight matrix, 2 total coef iterationsiterations

Instrument list: BETA_COMP_PRIN1_05_06 BETA_COMP_PRIN2_05_Instrument list: BETA_COMP_PRIN1_05_06 BETA_COMP_PRIN2_05_

06 BETA_COMP_PRIN3_05_06 BETA_COMP_PRIN4_05_0606 BETA_COMP_PRIN3_05_06 BETA_COMP_PRIN4_05_06

BETA_COMP_PRIN5_05_06BETA_COMP_PRIN5_05_06

VariableVariable CoefficieCoefficientnt

Std. Std. ErrErroror

t-Statistict-Statistic Prob. Prob.

CC 0.0003090.000309 0.000180.0001844

1.6740791.674079 0.19270.1927

BETA_COMP_PRINBETA_COMP_PRIN1_05_061_05_06

4.2830854.283085 0.201840.2018477

21.2194521.21945 0.00020.0002

BETA_COMP_PRINBETA_COMP_PRIN2_05_062_05_06

--0.80.8

7917915858

0.071270.0712766

-12.33456-12.33456 0.00110.0011

BETA_COMP_PRINBETA_COMP_PRIN3_05_063_05_06

0.1954350.195435 0.035900.0359022

5.4435655.443565 0.01220.0122

BETA_COMP_PRINBETA_COMP_PRIN4_05_064_05_06

--1.31.3

9179173030

0.093100.0931000

-14.94884-14.94884 0.00060.0006

BETA_COMP_PRINBETA_COMP_PRIN5_05_065_05_06

--0.80.8

3493498989

0.047840.0478477

-17.45122-17.45122 0.00040.0004

R-squaredR-squared 0.9926820.992682 Mean dependent Mean dependent varvar

0.006390.0063977

Adjusted R-Adjusted R-squaredsquared

0.9804850.980485 S.D. dependent varS.D. dependent var 0.007140.0071400

S.E. of regressionS.E. of regression 0.0009970.000997 Sum squared residSum squared resid 2.98E-062.98E-06

Durbin-Watson Durbin-Watson statstat

2.2112692.211269 J-statisticJ-statistic 3.59E-253.59E-25

Cross-section regression 2001 - 2003 2005 - 2006

Slide 16Slide 16

Page 17: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Risk premiums associated with factors Risk premiums associated with factors

CCOMP_PRIN1γ̂COMP_PRIN2γ̂COMP_PRIN3γ̂COMP_PRIN4γ̂COMP_PRIN5γ̂

Interval 2001 – 2003Interval 2001 – 2003 0.0105910.010591 -0.359591-0.359591 -0.086729-0.086729 -0.064155-0.064155 -0.414972-0.414972 0.3152640.315264

Interval 2005 – 2006Interval 2005 – 2006 0.0003090.000309 4.2830854.283085 -0.879158-0.879158 0.1954350.195435 -1.391730-1.391730 -0.834989-0.834989

COMP_PRIN1γ̂ COMP_PRIN2γ̂ COMP_PRIN3γ̂C COMP_PRIN4γ̂ COMP_PRIN5γ̂

Slide 17Slide 17

Page 18: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Predictions using sensitivity coefficients estimated through Two-step cross sectional regression procedure

Root Mean Root Mean Squared Error Squared Error

StatisticStatistic

Success ratio Success ratio sign sign

predictionprediction

Portfolio 1Portfolio 1 0.0228274490.022827449 0.350.35

Portfolio 2Portfolio 2 0.054693490.05469349 0.40.4

Portfolio 3Portfolio 3 0.021521430.02152143 0.30.3

Portfolio 4Portfolio 4 0.0516321610.051632161 0.40.4

Portfolio 5Portfolio 5 0.0754669390.075466939 0.40.4

Portfolio 6Portfolio 6 0.085837620.08583762 0.750.75

Portfolio 7Portfolio 7 0.0627300210.062730021 0.650.65

Portfolio 8Portfolio 8 0.0810786540.081078654 0.70.7

Portfolio 9Portfolio 9 0.0756745440.075674544 0.650.65

Root Mean Root Mean Squared Error Squared Error

StatisticStatistic

Success ratio Success ratio sign sign

predictionprediction

Portfolio 1Portfolio 1 0.000285150.00028515 00

Portfolio 2Portfolio 2 8.81318E-338.81318E-33 00

Portfolio 3Portfolio 3 0.0560572930.056057293 0.20.2

Portfolio 4Portfolio 4 0.080002030.08000203 0.60.6

Portfolio 5Portfolio 5 0.1367080460.136708046 0.60.6

Portfolio 6Portfolio 6 0.1142733760.114273376 0.650.65

Portfolio 7Portfolio 7 0.1453403560.145340356 0.50.5

Portfolio 8Portfolio 8 0.0791471860.079147186 0.70.7

Portfolio 9Portfolio 9 0.0681766120.068176612 0.70.7

2001 - 2003 2005 - 2006

Slide 18Slide 18

Page 19: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

One-step system of non-linear One-step system of non-linear seemingly unrelated equationsseemingly unrelated equations

An alternative method to the two-step procedure of estimating risk premium for economic observed variables was introduced by McElroy, Burmeister and Wall (1985), who demonstrated that the APT model can be expressed as a system of non-linear seemingly unrelated equations, in which factor loading and risk premium are estimated in one single step. The APT model has to parts: the procedure that generates returns and another equation for expected returns. By substituting expected returns in the returns generating equation, is resulting a single equation for APT:

it

k

jjtij

k

jjtijtit FbbR

110

or by passing to the left side the risk-free rate of return, the last equation becomes:

it

k

jjtij

k

jjtijtit FbbR

110

The risk-free rate of return is known, and considered for the purpose of this paper equal to the return on average annual interest rate for all existent deposits.

Slide 19Slide 19

Page 20: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

VariableVariable CoeffCoeff Std ErrorStd Error T-StatT-Stat SignifSignif

A1A1 -0.003233423-0.003233423 0.0016933660.001693366 -1.90946-1.90946 0.056202130.05620213

A2A2 -3.613009384-3.613009384 1.941765531.94176553 -1.86068-1.86068 0.062789010.06278901

A3A3 -0.004907741-0.004907741 0.0017100680.001710068 -2.86991-2.86991 0.004105890.00410589

A4A4 00 00 00 00

A5A5 -0.012755416-0.012755416 0.0014263020.001426302 -8.943-8.943 00

A6A6 00 00 00 00

A7A7 0.000228760.00022876 0.0025081740.002508174 0.091210.09121 0.927329060.92732906

A8A8 00 00 00 00

A9A9 -0.005935964-0.005935964 0.0026164180.002616418 -2.26874-2.26874 0.023284350.02328435

A10A10 00 00 00 00

Sum of SquaredSum of SquaredResidualsResiduals

R-squaredR-squared

Portfolio 1Portfolio 1 4.39059471344.3905947134 -0.008163-0.008163

Portfolio 2Portfolio 2 2.38350460562.3835046056 0.0138090.013809

Portfolio 3Portfolio 3 3.34699064523.3469906452 -0.010630-0.010630

Portfolio 4Portfolio 4 7.30232588797.3023258879 0.0500900.050090

Portfolio 5Portfolio 5 8.52216513578.5221651357 0.0319920.031992

Portfolio 6Portfolio 6 10.68069044210.680690442 0.0633380.063338

Portfolio 7Portfolio 7 8.15808194358.1580819435 0.0740720.074072

Portfolio 8Portfolio 8 15.65461060415.654610604 0.0897750.089775

Portfolio 9Portfolio 9 10.29827337210.298273372 0.1082190.108219

Coefficient estimates for the system of equations in interval 2001 – 2003

IN-SAMPLE statistics in interval 2001 – 2003

A1,A3,A5,A7,A9 represent sensitivity coefficientsA2,A4,A6,A8,A10 represents risk premiums

Slide 20Slide 20

Page 21: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

VariableVariable CoeffCoeff Std ErrorStd Error T-StatT-Stat SignifSignif

A1A1 -0.000920055-0.000920055 0.0003415290.000341529 -2.69393-2.69393 0.007061480.00706148

A2A2 -1.323749029-1.323749029 1.2384206071.238420607 -1.0689-1.0689 0.285114280.28511428

A3A3 -0.00042569-0.00042569 0.0009843230.000984323 -0.43247-0.43247 0.665399950.66539995

A4A4 00 00 00 00

A5A5 0.0003807330.000380733 0.000937730.00093773 0.406020.40602 0.684730970.68473097

A6A6 00 00 00 00

A7A7 0.0001799020.000179902 0.0009744960.000974496 0.184610.18461 0.85353460.8535346

A8A8 00 00 00 00

A9A9 -0.001476967-0.001476967 0.0009532450.000953245 -1.54941-1.54941 0.121283450.12128345

A10A10 00 00 00 00

Sum of Squared Sum of Squared ResidualsResiduals

R-squaredR-squared

Portfolio 1Portfolio 1 0.25714389260.2571438926 0.0232010.023201

Portfolio 2Portfolio 2 0.24607956430.2460795643 0.0231420.023142

Portfolio 3Portfolio 3 4.95236770344.9523677034 -0.001896-0.001896

Portfolio 4Portfolio 4 5.32890698425.3289069842 0.0016060.001606

Portfolio 5Portfolio 5 7.94903939017.9490393901 0.0036670.003667

Portfolio 6Portfolio 6 10.34068653510.340686535 -0.000686-0.000686

Portfolio 7Portfolio 7 5.60341136535.6034113653 0.0029440.002944

Portfolio 8Portfolio 8 13.04726531613.047265316 0.0003740.000374

Portfolio 9Portfolio 9 13.62814270313.628142703 0.0003430.000343

Coefficient estimates for the system of equations in interval 2005 – 2006

IN-SAMPLE statistics in interval 2005 – 2006

A1,A3,A5,A7,A9 represent sensitivity coefficientsA2,A4,A6,A8,A10 represents risk premiums

Slide 21Slide 21

Page 22: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

As can be seen from the IN-SAMPLE estimation results, in interval 2005 – 2006, the model can not explain the variation in portfolio returns. As a consequence, the system of non-linear equations is tested without Portfolio 1 and 2, which show extremely small variations in portfolios returns.

VariableVariable CoeffCoeff Std ErrorStd Error T-StatT-Stat SignifSignif

A1A1 -0.002013038-0.002013038 0.0008093960.000809396 -2.48709-2.48709 0.012879350.01287935

A2A2 -4.070064364-4.070064364 2.0469493382.046949338 -1.98836-1.98836 0.046772310.04677231

A3A3 -0.027149089-0.027149089 0.0023327640.002332764 -11.63816-11.63816 00

A4A4 00 00 00 00

A5A5 -0.041252281-0.041252281 0.0022223440.002222344 -18.56251-18.56251 00

A6A6 00 00 00 00

A7A7 -0.001289061-0.001289061 0.0023094750.002309475 -0.55816-0.55816 0.576733760.57673376

A8A8 00 00 00 00

A9A9 0.0059122390.005912239 0.0022591130.002259113 2.617062.61706 0.008869020.00886902

A10A10 00 00 00 00

Sum of Squared Sum of Squared ResidualsResiduals

R-squaredR-squared

Portfolio 1Portfolio 1 -- --

Portfolio 2Portfolio 2 -- --

Portfolio 3Portfolio 3 5.68933820425.6893382042 -0.150990-0.150990

Portfolio 4Portfolio 4 5.93915622105.9391562210 -0.112727-0.112727

Portfolio 5Portfolio 5 6.64323113016.6432311301 0.1673370.167337

Portfolio 6Portfolio 6 7.62050016137.6205001613 0.2625510.262551

Portfolio 7Portfolio 7 4.36900147844.3690014784 0.2225910.222591

Portfolio 8Portfolio 8 8.21553218478.2155321847 0.3705610.370561

Portfolio 9Portfolio 9 8.20956420048.2095642004 0.3978090.397809

Coefficient estimates for the system of equations in interval 2005 – 2006

IN-SAMPLE statistics in interval 2005 – 2006

A1,A3,A5,A7,A9 represent sensitivity coefficientsA2,A4,A6,A8,A10 represents risk premiums

Slide 22Slide 22

Page 23: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Root Mean Root Mean Squared Squared

Error Error StatisticStatistic

Success Success ratio sign ratio sign predictionprediction

Portfolio 1Portfolio 1 0.0237114440.023711444 0.350.35

Portfolio 2Portfolio 2 0.0539070750.053907075 0.350.35

Portfolio 3Portfolio 3 0.0256717690.025671769 0.30.3

Portfolio 4Portfolio 4 0.0521313840.052131384 0.350.35

Portfolio 5Portfolio 5 0.0752775260.075277526 0.40.4

Portfolio 6Portfolio 6 0.0903345820.090334582 0.70.7

Portfolio 7Portfolio 7 0.0679064650.067906465 0.70.7

Portfolio 8Portfolio 8 0.0836518860.083651886 0.650.65

Portfolio 9Portfolio 9 0.0842274190.084227419 0.550.55

]0[*40.00593596-]0[*00.00022876]0[*60.01275541-

]0[*10.00490774-]84-3.6130093[*23-0.0032334][ˆ

543

211

ttt

ttjt

k

jjtijit

FFF

FFFbR

The prediction results in interval 2001 – 2003 (20 observations)

Slide 23Slide 23

Page 24: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Root Mean Root Mean Squared Squared

Error Error StatisticStatistic

Success Success ratio sign ratio sign predictionprediction

Portfolio 1Portfolio 1 0.0344536390.034453639 0.60.6

Portfolio 2Portfolio 2 0.0344536390.034453639 0.60.6

Portfolio 3Portfolio 3 0.0588905020.058890502 0.450.45

Portfolio 4Portfolio 4 0.07297320.0729732 0.550.55

Portfolio 5Portfolio 5 0.1436213210.143621321 0.350.35

Portfolio 6Portfolio 6 0.1476096190.147609619 0.550.55

Portfolio 7Portfolio 7 0.1553104450.155310445 0.90.9

Portfolio 8Portfolio 8 0.1175769680.117576968 0.650.65

Portfolio 9Portfolio 9 0.0983180520.098318052 0.50.5

OUT-OF-SAMPLE statistics in interval 2005 - 2006

The prediction results in interval 2005 – 2006 (20 observations)

Root Mean Root Mean Squared Squared

Error Error StatisticStatistic

Success Success ratio sign ratio sign predictionprediction

Portfolio 1Portfolio 1 -- --

Portfolio 2Portfolio 2 -- --

Portfolio 3Portfolio 3 0.0679476630.067947663 0.40.4

Portfolio 4Portfolio 4 0.074413220.07441322 0.550.55

Portfolio 5Portfolio 5 0.1461706250.146170625 0.650.65

Portfolio 6Portfolio 6 0.1278093480.127809348 0.750.75

Portfolio 7Portfolio 7 0.1523254920.152325492 0.60.6

Portfolio 8Portfolio 8 0.1022031220.102203122 0.750.75

Portfolio 9Portfolio 9 0.0768142540.076814254 0.70.7

OUT-OF-SAMPLE statistics in interval 2005 – 2006

in the absence of Portfolio 1 and 2

Slide 24Slide 24

Page 25: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Feedforward Artificial Neural Feedforward Artificial Neural NetworksNetworks

•Network architecture. The feedforward neural network has a hidden layer, fully connected. The number of neurons at the input layer is 5 (corresponding to the 5 principal components) and a neuron on the output layer (representing the portfolio return). The number of neurons in the hidden layer is 15, a number set as a result of many tests with different number of neurons on the hidden layer.•Gradient descent terms. The BFGS (Boyden-Fletcher-Goldfarb-Shanno) algorithm approximates

1nH at step n based on the change in gradient 1 nn JJ

, relative to the change in the parameters 1 nn

. The epoch is kept always equal to one, meaning that the weights are updated after each presentation of a training pattern. This is the “on-line” or “stochastic” version of the BFGS algorithm, as opposed to the “batch” version where the weights are updated after the gradients have accumulated over the whole training set.•Transfer function, cost function and initial conditions. The transfer function is the logsigmoid function. The cost function used is the sum of squared differences between actual and estimated values. The initial conditions do not change through the training and prediction process.

Part IPart I

Slide 25Slide 25

Page 26: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Feedforward Artificial Neural Feedforward Artificial Neural Networks Networks

Part IIPart II

Mathematically the feedforward neural network can be described by the following equations:

K

ktkkt

tktk

I

itiikktk

LY

lL

Cl

1,0

,,

1,,0,,

)exp(1

1

where we have I=5 input variables and K=15 neurons in the hidden layer.

Slide 26Slide 26

Page 27: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

Elman Recurrent Artificial Neural Elman Recurrent Artificial Neural Networks Networks

K

ktkkt

tktk

I

i

K

ktkktiikktk

LY

lL

lCl

1,0

,,

1 11,,,0,,

)exp(1

1

Elman Recurrent Neural Network allow neurons in the hidden layer to depend not only on independent variables Ck at moment t, but also on their own lags. A “memory” effect is created in the neuron structure, similar to the moving average (MA) process in time-series analysis.The mathematical representation of the Elman Recurrent Network can be illustrated as follows:

Slide 27Slide 27

Page 28: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

IN-SAMPLE estimation results for the interval IN-SAMPLE estimation results for the interval 2001 – 20032001 – 2003

Two-step cross sectionalTwo-step cross sectionalregression procedureregression procedure

One-step system of One-step system of non-linear seeminglynon-linear seemingly unrelated equationsunrelated equations

Feedforward ArtificialFeedforward Artificial Neural NetworksNeural Networks

Elman Recurrent Elman Recurrent Neural NetworksNeural Networks

Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum of Sum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum of Sum of Squared Squared ResidualsResiduals

R-squaredR-squared

Portfolio 1Portfolio 1 3.9925503.992550 0.0016660.001666 4.39059471344.3905947134 -0.008163-0.008163 3.74883.7488 0.0440700.044070 3.7571743.757174 0.0651920.065192

Portfolio 2Portfolio 2 2.2615202.261520 0.0117030.011703 2.38350460562.3835046056 0.0138090.013809 2.163952.16395 0.0384020.038402 2.042912.04291 0.0914780.091478

Portfolio 3Portfolio 3 3.1496523.149652 0.0061290.006129 3.34699064523.3469906452 -0.010630-0.010630 3.00613.0061 0.0388040.038804 2.53552.5355 0.182610.18261

Portfolio 4Portfolio 4 7.1974177.197417 0.0357870.035787 7.30232588797.3023258879 0.0500900.050090 6.97446.9744 0.0606920.060692 6.5201856.520185 0.1211920.121192

Portfolio 5Portfolio 5 8.2267388.226738 0.0252150.025215 8.52216513578.5221651357 0.0319920.031992 7.96317.9631 0.0564210.056421 7.691977.69197 0.0954570.095457

Portfolio 6Portfolio 6 10.1233110.12331 0.0709380.070938 10.68069044210.680690442 0.0633380.063338 9.71649.7164 0.0999060.099906 9.06189.0618 0.143750.14375

Portfolio 7Portfolio 7 7.9189617.918961 0.0722610.072261 8.15808194358.1580819435 0.0740720.074072 7.44077.4407 0.113690.11369 7.280927.28092 0.1321260.132126

Portfolio 8Portfolio 8 13.2921513.29215 0.2077860.207786 15.65461060415.654610604 0.0897750.089775 12.61712.617 0.248910.24891 12.15812.158 0.267250.26725

Portfolio 9Portfolio 9 8.2448308.244830 0.2647190.264719 10.29827337210.298273372 0.1082190.108219 7.81037.8103 0.304980.30498 7.27667.2766 0.34050.3405

Slide 28Slide 28

Page 29: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

OUT-OF-SAMPLE prediction results for the OUT-OF-SAMPLE prediction results for the interval 2001 – 2003interval 2001 – 2003

Two-step cross sectionalTwo-step cross sectional regression procedureregression procedure

One-step system of One-step system of non-linear seeminglynon-linear seemingly unrelated equationsunrelated equations

Feedforward ArtificialFeedforward Artificial Neural NetworksNeural Networks

Elman Artificial Elman Artificial Neural NetworksNeural Networks

Root MeanRoot Mean Squared Squared

Error Error StatisticStatistic

SuccessSuccess ratio sign ratio sign predictionprediction

Root MeanRoot Mean Squared Squared

Error Error StatisticStatistic

SuccessSuccess ratio sign ratio sign predictionprediction

Root MeanRoot Mean Squared Squared

Error Error StatisticStatistic

Success Success ratio sign ratio sign predictionprediction

Root Mean Root Mean Squared Squared

Error Error StatisticStatistic

Success Success ratio sign ratio sign predictionprediction

Portfolio 1Portfolio 1 0.0234813640.023481364 0.40.4 0.0237114440.023711444 0.350.35 0.022191020.02219102 0.350.35 0.022018450.02201845 0.350.35

Portfolio 2Portfolio 2 0.0555783770.055578377 0.250.25 0.0539070750.053907075 0.350.35 0.054066760.05406676 0.40.4 0.052588170.05258817 0.40.4

Portfolio 3Portfolio 3 0.0220714540.022071454 0.350.35 0.0256717690.025671769 0.30.3 0.020514260.02051426 0.30.3 0.017154300.01715430 0.30.3

Portfolio 4Portfolio 4 0.0519028290.051902829 0.40.4 0.0521313840.052131384 0.350.35 0.049646920.04964692 0.350.35 0.046418260.04641826 0.650.65

Portfolio 5Portfolio 5 0.0740378620.074037862 0.30.3 0.0752775260.075277526 0.40.4 0.07364780.0736478 0.250.25 0.069453710.06945371 0.50.5

Portfolio 6Portfolio 6 0.0868231890.086823189 0.70.7 0.0903345820.090334582 0.70.7 0.087498570.08749857 0.70.7 0.086122010.08612201 0.750.75

Portfolio 7Portfolio 7 0.0626374050.062637405 0.650.65 0.0679064650.067906465 0.70.7 0.05721320.0572132 0.650.65 0.0606796520.060679652 0.650.65

Portfolio 8Portfolio 8 0.0807246350.080724635 0.650.65 0.0836518860.083651886 0.650.65 0.083339670.08333967 0.70.7 0.0807403240.080740324 0.750.75

Portfolio 9Portfolio 9 0.0744830770.074483077 0.550.55 0.0842274190.084227419 0.550.55 0.07411140.0741114 0.60.6 0.0693642560.069364256 0.650.65

Slide 29Slide 29

Page 30: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

IN-SAMPLE estimation results for the interval IN-SAMPLE estimation results for the interval 2005 – 20062005 – 2006

Two-step crossTwo-step cross sectional regressionsectional regression

procedureprocedure

One-step system ofOne-step system of non-linear seeminglynon-linear seemingly unrelated equationsunrelated equations

One-step system of One-step system of non-linear seeminglynon-linear seemingly

unrelated equations (*)unrelated equations (*)

FeedforwardFeedforward Artificial NeuralArtificial Neural

NetworksNetworks

Elman ArtificialElman Artificial Neural NetworksNeural Networks

Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum ofSum of Squared Squared ResidualsResiduals

R-squaredR-squared Sum of Sum of Squared Squared ResidualsResiduals

R-squaredR-squared

Portfolio 1Portfolio 1 0.0120440.012044 0.0024910.002491 0.2571438920.25714389266

0.0232010.023201 -- -- 0.011340.01134 0.0033170.003317 0.0113360.011336 0.00473110.0047311

Portfolio 2Portfolio 2 3.36E-613.36E-61 -- 0.2460795640.24607956433

0.0231420.023142 -- -- -- -- -- --

Portfolio 3Portfolio 3 4.4108944.410894 0.0344720.034472 4.9523677034.95236770344

-0.001896-0.001896 5.68933820425.6893382042 -0.150990-0.150990 4.35654.3565 0.0385840.038584 4.029974974.02997497 0.0932250.093225

Portfolio 4Portfolio 4 4.9848774.984877 0.0346410.034641 5.3289069845.32890698422

0.0016060.001606 5.93915622105.9391562210 -0.112727-0.112727 4.7649774.764977 0.0625260.062526 4.11054.1105 0.144980.14498

Portfolio 5Portfolio 5 6.3612336.361233 0.1792660.179266 7.9490393907.94903939011

0.0036670.003667 6.64323113016.6432311301 0.1673370.167337 6.12086.1208 0.198560.19856 5.86335.8633 0.210550.21055

Portfolio 6Portfolio 6 7.1539757.153975 0.2958820.295882 10.3406865310.3406865355

-0.000686-0.000686 7.62050016137.6205001613 0.2625510.262551 6.838266.83826 0.294760.29476 6.39476.3947 0.340060.34006

Portfolio 7Portfolio 7 4.0815564.081556 0.2374970.237497 5.6034113655.60341136533

0.0029440.002944 4.36900147844.3690014784 0.2225910.222591 3.75513.7551 0.266230.26623 3.618213763.61821376 0.3022570.302257

Portfolio 8Portfolio 8 6.2450826.245082 0.5066260.506626 13.0472653113.0472653166

0.0003740.000374 8.21553218478.2155321847 0.3705610.370561 5.96315.9631 0.554520.55452 5.6065.606 0.570020.57002

Portfolio 9Portfolio 9 5.4871605.487160 0.5902970.590297 13.6281427013.6281427033

0.0003430.000343 8.20956420048.2095642004 0.3978090.397809 2.939412.93941 0.767710.76771 4.823964744.82396474 0.6338030.633803(*)Estimation results in interval 2005 – 2006 using One-step System of Non-linear Seemingly Unrelated Equations in the absence of Portfolios 1 and 2

Slide 30Slide 30

Page 31: Estimating and Predicting Stock Returns Using Artificial Neural Networks Dissertation Paper BUCHAREST ACADEMY OF ECONOMIC STUDIES DOCTORAL SCHOOL OF FINANCE

OUT-OF-SAMPLE prediction results for OUT-OF-SAMPLE prediction results for the interval 2005 – 2006the interval 2005 – 2006

(*) Prediction results in interval 2005 – 2006 using One-step System of Non-linear Seemingly Unrelated Equations in the absence of Portfolios 1 and 2

Slide 31Slide 31

Two-step cross sectional regression

procedure

One-step system of non-linear

seemingly unrelated equations

One-step system of non-linear seemingly unrelated equations

(*)

Feedforward Artificial Neural

Networks

Elman Artificial Neural Networks

Root Mean Squared

Error Statistic

Success ratio sign prediction

Root Mean Squared

Error Statistic

Success ratio sign

prediction

Root Mean Squared

Error Statistic

Success ratio sign prediction

Root Mean Squared

Error Statistic

Success ratio sign

prediction

Root Mean Squared

Error Statistic

Success ratio sign prediction

Portfolio 1 0.000294522 0 0.034453639 0.6 - - 0 0 0.00077611 0

Portfolio 2 1.24522E-32 0 0.034453639 0.6 - - - - - -

Portfolio 3 0.055832265 0.15 0.058890502 0.45 0.067947663 0.4 0.05757300 0.2 0.05274362 0.3

Portfolio 4 0.079836677 0.5 0.0729732 0.55 0.07441322 0.55 0.08316120 0.45 0.08024026 0.55

Portfolio 5 0.136755982 0.6 0.143621321 0.35 0.146170625 0.65 0.13940947 0.65 0.13969073 0.5

Portfolio 6 0.113841636 0.35 0.147609619 0.55 0.127809348 0.75 0.11709767 0.65 0.11818629 0.55

Portfolio 7 0.145922651 0.45 0.155310445 0.9 0.152325492 0.6 0.14058805 0.6 0.14406041 0.65

Portfolio 8 0.079475156 0.7 0.117576968 0.65 0.102203122 0.75 0.07137226 0.7 0.08297590 0.7

Portfolio 9 0.06808954 0.7 0.098318052 0.5 0.076814254 0.7 0.05801578 0.8 0.07260961 0.8