Download CP5_propensity scoring to genr matched pairs of DRIP vs non

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Regression analysis wikipedia , lookup

Transcript
FIN 938 Investments Seminar
CP5 – Propensity Scoring: Analyzing Matched Pairs of DRIP Firms vs Non-DRIP Firms
Analyze the data provided in the following library on my CRMDA account:
libname CP5
'R:\users\pkoch\PK_Space\FIN 938_computer
problems\CP5_Propensity Scoring to Genr Matched Pairs_DRIPs';
This folder contains a number of datasets, plus sas code to generate and analyze a subset of
matched pairs of firms with DRIPs and without DRIPs each quarter over the sample period
covering the years, 2010 - 2012, requiring that the matched pairs of firms be from the same
industry, and be based on similarities in certain firm characteristics.
The key variable is the quarterly benchmark-adjusted abnormal return on the dividend pay date
(bm_ar0 = AR(0)), for all dividend-paying U.S. firms over the period, 2010 - 2012. We want to
know if this variable behaves differently for DRIP firms versus ‘comparable’ non-DRIP firms.
Explanatory variables:
1) DRIP = 1, if the firm has a company-sponsored dividend reinvestment plan, or 0 otherwise.
The list of DRIP firms is available once per year for the years, 2096 - 2012, from the
American Association of Individual Investors (AAII);
2) mktcap = the firm’s daily market capitalization on day -10, two weeks before the firm’s
quarterly dividend pay date (on day 0), taken from CRSP;
3) lnsize = ln_size = ln(size) = the natural log of the firm’s daily market capitalization on
day -10, two weeks before the dividend pay date (on day 0);
4) yld = Div_Yield = firm’s dividend yield, computed as the cash dividend amount divided
by the firm’s stock price on day -10, where the dividend amount is taken from CRSP;
5) Pct_INST = IO = percent of shares outstanding owned by institutional investors for the
current quarter, taken from Thompson’s 13F filings;
6) LPct_INST = percent of shares outstanding owned by institutional investors for the
previous quarter, taken from Thompson’s 13F filings;
7) spread = normal_spread = daily percentage closing bid-ask spread, as a percent of the daily
closing price, taken on day -10, from CRSP;
8) log_hilo = normal_log_hilo = intraday stock return volatility measured as the natural log
of the ratio of the daily high and low prices, on day -10, taken from CRSP;
9) broker_non = proportion of total shares outstanding for the firm that are classified as broker
non-votes in quarterly proxy contests to elect members of the firm’s board;
10) PART(1) = Participation rate in the firm’s DRIP = 1 – Pct_INST – Broker_non_votes.
1
Your Assignment:
A. Generate a dataset containing a subset of all U.S. dividend-paying stocks each quarter, that
includes a set of matched pairs of firms with DRIPs and without DRIPs, that are from the same
industry (based on the Fama-French 5-industry classifications), and that further have
‘comparable’ values of the following firm characteristics: lnsize, yld, and spread.
B. Compute and present a table that compares the mean values of the following characteristics,
for the matched sample of firms with DRIPs versus comparable firms without DRIPs:
bm_ar0, mktcap, div_yield, normal_spread, normal_log_hilo, part1, IO, and broker_non.
C. Analyze the association between the difference in the AR(0) across matched pairs of firms
with and without DRIPs (d_bm_ar0), and the difference in the DRIP participation rates across
these firms (d_part), with and without controlling for the differences in other firm characteristics
(i.e., d_div_yield, d_ln_size, d_normal_spread, and d_log_hilo). This analysis requires
estimation of the following panel regression model:
Model: bm_ar0 = dpart d_div_yield d_ln_size d_normal_spread d_log_hilo).
You should use the following two techniques to estimate this panel regression:
the Fama-MacBeth approach and a panel with clustered standard errors on the matched pair (idt)
and the quarter (n). Compare the results for these two approaches, and summarize the
implications of the results.
2