LongShort Trading Strategy

1
Long/Short Trading Strategy

• Cams Crazies
• Global Asset Allocation
• February 2005

2
Agenda

• Methodology
• Factors
• UniVariate Models
• Book to Price
• Dividend Yield
• FY2 Earnings Yield to Growth
• BiVariate Model
• Scoring Model
• Why Results Might Be Wrong
• Follow-up Research
• Conclusion

3
Methodology

• Objective To develop a quantitative long/short
  model that generates positive and consistent
  returns with no market exposure (beta zero).
• Universe For the sake of liquidity and
  availability of historical data, we limited our
  screening universe to the 500 largest market
  capitalization companies listed in the U.S.
• Rebalancing To limit turnover and transaction
  expenses, we resort and rebalance annually.
• Weights For simplicity and prevention of outlier
  performance, we chose an equal weight strategy,
  both in the number of stocks per bucket and in
  the amount invested or shorted per stock.

4
Factors

• Book to Price
• Earnings Yield (FY1/FY2)
• FY1/FY2 Earnings Yield to Growth
• Earnings Growth
• Dividend Yield
• 3 Year EPS Growth

5
Factors Book to Price
Go long stocks with high B2P ratios (low Price to
Book) and short stocks with low B2P ratios.
Average annual return of 9 and .53 Sharpe ratio.
High volatility of returns (48 and -23).
The screen did not perform well during the
1998-1999 valuation bubble, producing -21 and
-11 returns, respectively.
6
Factors Dividend Yield
Go long high dividend yield stocks short low/no
dividend yield stocks. Some sector weight and
value vs. growth concerns. Best screen 16.7
beta-neutral average returns, 1.11 sharpe ratio,
lowest turnover of all factors. Alphas for
fractiles 1 and 5 11.6 and -7.7, respectively.
Most consistent results fractile 1 was top
performer all but 2 years. Bottom fractile
nearly always 4 or 5. Consecutive losses in
1998 and 1999, but positive returns in all other
years.
7
Factors FY2 Earnings Yield to Growth
Forward looking uses fiscal earnings estimates
for two years out scaled by consensus long term
growth expectations. Go long high EY to growth
fractile, short low EY to growth fractile.
A beta neutral long/short strategy resulted in an
average return of 10 with a range of -11 to
37. High Sharpe ratio of .78. There were four
years of negative returns over the 20-year time
period. Consecutive losses in 1998 and 1999.
8
BiVariate Model

• Factor 1 Dividend Yield fundamental data
• Factor 2 FY2 Earnings Yield to Growth
  expectational data
• Two sorts produced 25 fractiles with 20
  companies each long top fractile, short bottom
  
• Trouble sorting in later years damages
  credibility of data.
• Average beta-weighted return of 29 and sharpe
  ratio of 1.58 are exceptionally high.
• Returns in bubble years range from 32-58. 2004
  only year of negative returns (-8).

9
Scoring Model
Selected Book to Price, Dividend Yield, Earnings
Yield (FY2) and Earnings Yield (FY2) to growth.
These factors combine both fundamental and
expectational variables.
We subjectively chose scores, ranging from -5 to
5, for fractiles 1 and 5. Highest possible Score
is 12, lowest possible score is -9.
Beta-neutral average returns of 12, sharpe ratio
of .81 not as good as stand-alone dividend yield.
10
Why Might Results Be Wrong?

• Model depends on FactSet both the accuracy of
  the data and Alpha Tester program.
• Data could be subject to
• errors
• lags
• survivorship bias
• outliers
• high minus low
• sector bets

11
Follow-Up Research

• Detailed review of the accuracy of the FactSet
  historical data and the Alpha Tester model.
• Suspicious beta screen
• Model Specific
• more frequent rebalancing
• a larger universe of companies/ international
  markets
• a more exact estimation of the impact of
  turnover, management fees, and short sale
  restrictions
• optimization of weights in Scoring model
• ex-ante application of the model on a real time
  basis

12
Beta as a Factor?
Persistent Characteristic Long low beta
portfolio, Short high beta portfolio.
Decided to run screen based on low-to-high Beta
Fractiles. Screen generated about 8.0 of Alpha
although heat map suggests inconsistency.
13
Conclusion

• Results were very impressive
• Further research is needed to gain confidence in
  models
• How long will excess returns persist before
  identified and competed away?
• That said, we cant help but wonder

What if were right??