1. Determinants of Capital Structure Choice: A Structural Equation Modeling Approach Cheng F. Lee Distinguished Professor of Finance Rutgers, The State University of New Jersey Editor of Review of Quantitative Finance and Accounting and Review of Pacific Basin Financial Markets and Policies The 15th Annual Conference on PBFEAM at Ho Chi Minh City, Vietnam

2. OUTLINE I. Introduction II. Measures and Determinants of Capital Structure III. Sample IV. Methodology V. Empirical Results IV. Conclusion

3. I. Introduction A. History of Finance B. Theoretical Framework of Finance a. Classical Theory b. New classical theory c. CAPM and APT d. Options and Futures Theory C. Policy Framework of Finance a. Investment Policy b. Financial Policy c. Dividend Policyd. Production Policy D. Accounting Approach to Determine Capital Structure a. Static Ratio Analysis b. Dynamic Ratio Analysis E. Finance Approach to Determine Capital Structure a. Traditional Approach b. Option Approach

4. II. Measures and Determinants of Capital Structure A. Growth B. Uniqueness C. Non-Debt Tax Shields D. Collateral Value of Assets E. Profitability F. Volatility G. Industry Classification

5. III. Sample

6. Table 1 Descriptive Statistics for the Pooled Sample during 1988-2003

7. Table 2 Pearson Correlation Coefficients for the Pooled Sample Table 2 Pearson Correlation Coefficients for the Pooled Sample during 1988-2003, N = 13,387

8. IV. Methodology A. MIMIC Model B. Estimation Criterion C. Model Fit Evaluation

9. Table 3 Constructs and Indicators in Titman and Wessels (1988) Model

10. Figure 1. Path Diagram of a Simplified MIMIC A. MIMIC Model

11. (1)  =      +  Y =  y    (2) X =  x   ,

12. Let  =0, X ,  x =I, and  =0, the full structural equation model becomes a MIMIC model  =  X +  Y =  y   , where  is a (m x 1) vector of endogenous variables with zeros on the diagonal;  is a (n x 1) vector of exogenous variables;  is a (m x 1) vector of errors in equation.

13. The latent variable  is linearly determined by a set of observable exogenous causes, X = (X 1, X 2, …, X q )’, and a disturbance . In matrix form  =  X +  or in equation form  =  ’X +  =  1 X 1 +  2 X 2 + …+  q X q + .

14. The latent variable, in turn, linearly determines a set of observable endogenous indicators, Y = (Y 1, Y 2, …, Y p )’ and a corresponding set of disturbance,  = (  1,  2, …,  p )’. In matrix form Y =  y   . In equation form Y 1 = 1  +  1 Y 2 = 2  +  2 … Y p = p  +  p.

15. The disturbances are mutually independent due to the fact that correlations of Y’s are already accounted for by their common factor or so-called latent variable, . For convenience, all variables are taken to have expectation zero. That is, the mean value of each variable is subtracted from each variable value. Thus, E(   ’) = 0’, E(  2 ) = , E(  ’) =  , where  is a (p x p) diagonal matrix with the vector of variances of the  ’s, , displayed on the diagonal.

16. The equations can be combined to yield a reduced form Y =  y    =  y (  ’X +  ) +  = (  y  ’) X +  y  +  =  ’ X + (  y  +  ) =  ’ X + z, where  =  y  ’ is the reduced form coefficient matrix; z =  y  +  is the reduced form disturbance vector.

17. The disturbance vector has covariance matrix Cov(z) =  = E(zz’) = E[(  y  +  )(  y  +  )’] =  y  y ’  +   Where  = Var(  ) and  is diagonal covariance matrix of .

18. B. Estimation Criterion F = log ||  || + tr(S  -1 ) – log||S|| - (p + q), Where  is the population covariance matrix; S is the model-implied covariance matrix; p is the number of exogenous observable variables; q is the number of endogenous observable variables.

19. V. Empirical Result

20. Table 4 Constructs, Causes and Effects in MIMIC Model

21. Table 5 Goodness-of-Fit Measures

22. Table 6 Completely Standardized Loadings

23. Table 6 (Cont ’ d) Completely Standardized Loadings

24. Table 7 Significance of Unstandardized Total Effect of Determinants of Capital Structure

25. Table 8 Signs of Total Effect of Determinants of Capital Structure

26. Figure 2 Relative Impact of Determinants of Capital Structure

27. Table 9 Squared Multiple Correlations

28. Table 10 Comparison of the Empirical Results

29. VI. Conclusion VI. Conclusion A. The Results Obtained from MIMIC Model Performed Better than those from LISREL Model B. Growth, Uniqueness, Non-Debt Tax Shields, Collateral Value of Assets, Profitability, Volatility, and Classification are the Six Important Characteristics for Determining the Capital Structure of a Firm C. In Practice, Capital Structure Information can be used to Estimate Financial Z-score, Cost of Capital Estimation. In addition, Capital Structure is Important for Performing Credit Risk Analysis. D. Capital Structure Theories can also be used to do Macro-Finance and Economic Policy Research E. Investment, Financing, and Dividend and Production Policies are Important in Corporate Governance Research.

30. Appendix A. Path Diagram Implied in Titman and Wessels (1988) Model

31. Appendix B: Completely Standardized Total Effect of Determinants of Capital Structure