Other Methods

Other Methods in Finance

There many important methods in (financial) empirical economics but I do not regularly work with all these methods. Some of them are important for common student thesis work. I make some basic comments that I collected from the literature on the use of some of these methods and the complications associated with them. Note that many of the lecture notes available on websites only provide a brief introduction and do not deal with many of the empirical difficulties that must be addressed in serious academic work.

Event studies

The typical event study in finance attempts to measure the effect of an economic event on the stock market value of a firm. Of course, the methodology can be applied to other questions as well. Simulation studies show that the basic type of event study is quite reliable and robust to alternative assumptions and problems when: 1) the event date is clearly specified (no uncertainty, little information leakage), and 2) abnormal returns are examined over short periods (days rather than months). On the other hand, however, event studies for long-horizon abnormal returns (usually taken to be more than 12 months) are problematic and require very careful consideration.

Two basic approaches are suggested:

* Event-time analysis: buy-hold abnormal returns, using benchmark returns based on multiple-factor models or characteristics matched firms.

* Calendar-time analysis: calendar-time returns for a portfolio of event firms (equal-weighted or value-weighted, firms having experienced events within the previous T periods), evaluated for abnormal returns using Jensen's alpha estimated from a multiple-factor model, or abnormal returns estimated from a portfolio of characteristics matched firms.

The basic methodology of event studies is readily available, see the references listed below. I mention some key issues to watch out for:

* measuring abnormal returns (AR): choice of benchmark model, estimated pre-event or post-event

  • (Ri - Rm)
    • Note: using the market-adjusted return is suspect, because it does not adjust for basic CAPM beta risk
  • (Ri - α - βRm)
    • Note: using the market-model is suspect, because it assumes that the risk-free interest rate included in α (α = α' + (1-β) Rf) is constant whereas market returns are assumed to change
  • (Ri - Rf - α - β(Rm-Rf))
    • Note: using the CAPM-model is suspect, because β-risk is known to be an incomplete adjustment for risk (see for example, Fama-French 3 factor model). The α is estimated from pre-event data and captures unexplained special circumstances that will automatically cause a bias in post-event returns
  • (Ri - α - Σ βj Fj) multi-factor models, for example Fama-French 3-factor, Carhart 4-factor or APT. More recent empirical studies have suggested important effects of liquidity and downside risk (skewness).
    • Note: same issues exist as with the basic CAPM model, but hopefully to a lesser degree
  • Matched-firms return.
  • Note: sensitive to the matching factors selecting by the researcher

* measuring cumulative abnormal returns (CAR or BHAR): rebalancing (EW) or buy-hold (VW) portfolio assumption. Rebalancing introduces a positive rebalancing bias in longer horizon returns. Rebalancing significantly increases the amount of trading costs in the implied investment strategy.

* statistical significance (standardized AR and CAR or SAR and SCAR): cross-section differences and time-varying volatility of returns: heteroskedasticity. Variances of the return observations across events are not equal and variances over time are not constant and therefore require appropriate adjustments.

* statistical significance: independent or related events. Events are usually not independent, but for example events in economics tend to cluster in specific time periods. This affects the standard error of estimates and needs appropriate adjustment.

* statistical significance: besides (standardized) t-tests, nonparametric rank statistics and sign statistics provide more robust tests in the face of non-normality of returns. Returns in financial studies are known to be subject to skewness and fat tails or kurtosis, invalidating the normality assumption underlying the usual test statistics. Bootstrap test procedures are also an alternative and there is also a skewness adjusted t-statistic (see Lyon, Barber and Tsai, 1999).

* explaining the differences in (C)AR: 2nd stage regression analysis. A regression to find find explanatory variables for event returns.

A collection of selected references to the methodology of event studies.

* Eventus User's Guide. Appendix A: Technical Reference http://www.eventstudy.com/Eventus-Guide-8-Public.pdf

Particularly the availability and definition of alternative test statistics.

* Corrado and Truong (2008), Conducting event studies with Asia-Pacific security market data, Pacific-Basin Finance Journal vol.16 November

* Kothari and Warner (2006), Econometrics of event studies, Chapter 1 in Eckbo (ed.) Handbook of Corporate Finance: Empirical Corporate Finance, Elsevier/North Holland.

http://mba.tuck.dartmouth.edu/pages/faculty/espen.eckbo/PDFs/Handbookpdf/CH1-EventStudies.pdf

* Campbell, Lo and MacKinlay (1997), Event-study analysis, Chapter 4 in The Econometrics of Financial Markets, Princeton Univ. Press: 149-80.

* Blinder (1998), The event study methodology since 1969, Review of Quantitative Finance and Accounting, vol.11 (September) http://www.springerlink.com/content/x60613477u424882/

* M.J. Seiler (2004), Event Studies, Chapter 13 in Performing Financial Studies: A Methodological Cookbook. Pearson Prentice Hall.

Practical explanation of doing manual event studies in Excel. http://wps.prenhall.com/bp_seiler_pfs_1/

No empirical method is perfect. Several studies point to methodological problems and weaknesses (and suggestions for improvement) that have to be taken into account when doing empirical research with event studies.

* Lyon, Barber and Tsai (1999), Improved methods for tests of long-run abnormal stock returns, Journal of Finance, vol.54 (February) http://mc1litvip.jstor.org/stable/222413

* Fama (1998), Market efficiency, long-term returns, and behavioral finance, Journal of Financial Economics, vol.49 ( ): 283-306

* Ahern (2008), Sample selection and event study estimation, working paper. http://webuser.bus.umich.edu/kenahern/Ahern.SSESE.pdf

* McWilliams, Siegel and Teoh (1999), Issues in the use of the event study methodology, Organizational Research Methods, vol.2 (4): 340-65. http://orm.sagepub.com/cgi/content/refs/2/4/340

* Lease, Masulis and Page (1991), An investigation of market microstructure impacts on event study returns, Journal of Finance vol.46(4) September http://www.jstor.org/sici?sici=0022-1082%28199109%2946%3A4%3C1523%3AAIOMMI%3E2.0.CO%3B2-7&origin=repec

Web pages with additional material, lecture notes:

Note that many of the lecture notes only provide a brief introduction and do not deal with many of the empirical difficulties that must be addressed.

* http://www.eventstudy.com/

Commercial software package Eventus

* The Event Study Webpage

http://web.mit.edu/doncram/www/eventstudy.html

* Michael Lemmon, University of Utah: Finance 7870 Empirical Methods in Finance

http://home.business.utah.edu/finmll/fin787/slides/eventstudiesclm.pdf

* Seppo Pynnonen, University of Vaasa, TIL.305 Financial Econometrics

http://lipas.uwasa.fi/~sjp/Teaching/Efm/Lectures/fec4.pdf

* Gerald Dwyer, Dwyer Economics, lecture slides

http://www.dwyerecon.com/pdf/lecteven.pdf

* Paul Laux, University of Delaware: Notes on methodology of long run returns studies

http://www.buec.udel.edu/laux/LearnCurve/long_run_returns.pdf

Sorting and portfolio returns

A wealth of empirical evidence connects numerical factors associated with specific equities to the cross-section of expected returns. For example, firm characteristics and prior price history have been shown to be related to expected returns. The methodology is closely related to event study analysis, although the basic assumptions concerning the "sorting events" are different and event study analysis tends to focus on short-term rather than long-term (average) returns.

One important consideration in financial market studies is that the sorting event information (on which the portfolios will be formed) must be available in real time.

A collection of selected references to the methodology of sorting and return evaluation.

* Lo and MacKinlay (1990), When are contrarian profits due to stock market overreaction?, Review of Financial Studies, vol.3: 175-205

* Lo and MacKinlay (1990), Data-snooping biases in tests of financial asset pricing models, Review of Financial Studies, vol.3: 431–467.

* Berk (2000), Sorting out sorts, Journal of Finance, vol.55, 407-27.

* Liang (2000), Portfolio formation, measurement errors, and beta shifts: A random sampling approach, working paper. * Conrad, Cooper and Kaul (2003), Value versus glamour, Journal of Finance, vol.58: 1969-96

* Boynton and Oppenheimer (2006), Anomalies in stock market pricing: Problems in return measurements, Journal of Business, vol.79: 2617-31.

* Fan and Liu (2007), Sorting, firm characteristics, and time-varying risk: An econometric analysis, Journal of Financial Econometrics, vol.6: 49-86.

* Liu and Strong (2008), Biases in decomposing holding-period portfolio returns, Review of Financial Studies, vol.21: 2243-74.

Web pages with additional material, lecture notes:

*