Transcript 不同實證結果的爭論?
漢珍2008商學研究與數位資源 應用研討會 分量迴歸的介紹與應用 降息刺激經濟成長？ 如何增加公司利潤？ 漲時看勢，跌時看質？ How Many Years for One SSCI Publication? How Many Years for One SSCI Publication? 2004 2005 2006 2007 2008 SSCI 發表總篇數 1 1 0 2 7+1 IF>0.5 之 SSCI 發表總篇數 國科會評定之優良國際期刊 0 0 0 0 0 0 1 1 5 5 0.327 0.327 0 0.849 3.518 IF 總合 計量模型與方法的應用 • 分析並呈現資料的特性 • 解釋過去的經濟現象 • 預測未來 資料 • 橫斷面 (cross-section) 資料: • 時間序列 (time series) 資料: • 追蹤資料 (panel) 資料: 不同實證結果的爭論? • 計量模型失靈? • 計量學家失靈? • 實證資料不對? • 瞎子摸象？ 不同實證結果的爭論? • Certain puzzles of the relations between firm profitability and its determinants: – R&D expense – Size effect on the profitability – Debt ratio puzzle 不同實證結果的爭論? • R&D expense – Branch (1974); Grabowski and Mueller (1978); Reinhard (1985); Guerard, et al. (1987); Brown (1988); Kraft (1989): Chan, et al. (1990); Morbey and Reithner (1990); Sougiannis (1994); Deng, et al. (1999) and Schoenecker and Swanson (2002) 不同實證結果的爭論? • The effect of size on the profitability – Ferri and Jones (1979), Smith and Watts (1992), Panzar and Willig (1979), Eckard (1990) and Paul (2001) – Williamson (1967), Holmes, et al. (1991); Lever (1996); Chuang (1999); Pull (2003) 不同實證結果的爭論? • Capital structure (Debt ratio): – Kim (1978), Schneller (1980) and Bradley, et al. (1984) – Titman and Wessels (1988) and Baskin (1989) 計量方法的演變 Yi =β1+β2 X2i+…+ βk Xki+ ei , i =1,…,N • Specification linear model nonlinear model nonparametrics • Method least squares (least absolute deviation), QMLE, GMM, computationally intensive methods, etc. • Structure single equation multiple equations (simultaneous-equation model, VAR) Does the era of linearity and leastsquares come to an end? • Legendre: – Least squares: – Minimizing sum of squared errors • Boscovich: – Least absolute deviation – Minimizing sum of absolute errors • Key: “averaging” the errors Legendre Boscovich An Example: Simple Regression An Example: Multiple Regression Regression Line OLS and LAD yi x u i ' i min 1 u 1 ( y 2 i i min i 2 i 1 | u |1 | y i i x ) ' i i i x | ' i 平均數具有代表性嗎？ • 應用和計算平均數的前提條件是什麼？ – 同質性的分配 – 因為只有是同質性的分配，計算平均數才有意 義 The Perspective of Quantile Regression (QR) The Perspective of Quantile Regression (QR) How to Enhance Firm Profit? Motivations • Limitations of OLS and LAD methods – Central behaviors only – Conditional mean • Qunantitle Regression (QR) Model – Whole distribution – Conditional distribution Key Question • The impacts of the identified determinants of profitability performance on firms are consistent with different levels of firm’s profitability quantiles? Motivations • Certain puzzles of the relations between firm profitability and its determinants: – R&D expense – Size effect on the profitability – Debt ratio puzzle Motivations • Firm life cycle • Adizes (1988): – Business strategies and organizational structures of firms vary according to the problems faced at different life cycle stages of the organization Prior Studies on Life-cycle Theory • The concept of corporate life-cycle stage has generated considerable applied interest – Dodge et al. (1994) on the relationship between operation strategies and life cycles – Beldona et al. (1997) and Robinson (1998) on effects of operation strategies on performance at various firm life cycle stages – Kimberley and Miles (1980) and Dodge and Robins (1992) positing that organization structure reflects current life cycle stage – Adizes (1979); Miller and Friesen (1984); Alexander et al. (1993); and Maturi (1999) examining the relationship between CEO leadership styles and life cycle stages – Anthony and Ramesh (1992), Black (1998) and Jorion and Talmor (2001) testing the impact of life-cycle on corporate earnings. Key Limitation of Prior Studies • Segment sample companies into various subsets – Use criteria such as earnings and/or age – Apply traditional optimization techniques such as ordinary least squares (OLS) and least absolute deviation (LAD) to fit their subsets Key Limitation of Prior Studies • The analytical framework in these studies was based on unconditional distribution of firm samples • This form of “truncation of samples” may yield invalid results • As demonstrated by Heckman (1979), such methods often exhibit sample selection bias Key Advantage of QR • A valid alternative is the quantile regression framework, which segments the sample into subsets defined by conditioning covariates • Moreover, in comparison with the least square method, quantile regression offers a relatively rich description of the conditional mean for extreme cases in the samples Empirical Methods • No quantile models: OLS and LAD yi x ui ' i min 1 u 1 ( y 2 i i min i 2 i 1 | u |1 | y i i x ) ' i i i x | ' i Empirical Methods • Quantile regression model y i x ui ' i Quant ( y i xi ) inf y : Fi ( y x) x ' i Quant (ui xi ) 0 Empirical Methods • Quantile regression model min | u i i:ui 0 i: yi xi' 0 | (1 ) ui i:ubi 0 | yi xi' | ' ( 1 ) | y x i i i: yi xi' 0 Data • Sample S&P500 firms over the 10-year period from 1996 to 2005 were analyzed • Financial firms were excluded • Firms were also excluded from the analysis if the specified financial data were not available for the entire 10-year period • The final sample included 2,078 firm observations • All data were obtained from the Compustat database. Data • Return on equity (ROE) was selected as the proxy variable for firm profitability • Five determinants of profitability were recorded: – – – – – R&D expense Company size Debt ratio Total asset turnover Current ratio Empirical Results Table 1 Definitions of Dependent/Independent Variables Variables Dependent Variable ROE Independent Variables RD SIZE DR TAT CR Definitions Return on equity= Net income/ shareholder’s Equity (ROE reveals how much profit a company generates with the money shareholders have invested in the company) Per research and development expenses=R&D expenses/employee numbers (Discovering new knowledge about products, processes, and services, and then applying that knowledge to create new and improved products, processes, and services that fill market needs) Natural log of book value of total asset=Ln (total assets) Debt ratio=Total liabilities/total assets (Debt ratio shows how much the company relies on debt to finance assets) Total asset turnover=Net sales/total assets (TAT is a measure of how well assets are being used to produce revenue) Current ratio=Current assets/current liability (Current ratio is an indication of a company's ability to meet short-term debt obligations; the higher the ratio, the more liquid the company is) Conclusions and Directions for Future Research • A comparative analysis: QR, OLS and LAD estimates • The nonlinearities derived from conditional QR: – size, significance and sign • It should be noted that certain existing puzzles of the relations between firm profitability and its determinants could be satisfactorily accounted for in this study Conclusions and Directions for Future Research • Certain caveats – U.S. S&P 500 firms – Comparisons with models with other dynamic parameter designs – Proxy variable selection – Other determinants of firm profitability – Panel data Application #2: 降息刺激經濟成長？ Table 1 Definition of Dependent/Independent Variables Variables Definitions Dependent Variable GGDP Growth rates of real GDP Independent Variables GGOV SR Growth rates on government-consumption-expenditure/GDP ratios 3-month short term interest rates GIVT GEXT Growth rates on investment/GDP ratios Growth rates on export/GDP ratios GIMT Growth rates on import/GDP ratios Application #3 漲時看勢，跌時看質？