Presentation by:

1
Granite Investments
Global Asset Allocation

• Presentation by
• Austin Applegate
• Michael Cormier
• Paul Hodulik
• Carl Nordberg
• Nikki Zadikoff

February, 26 2004
2
Agenda
Agenda

• Introduction
• Methodology
• Factors
• 1-Year EPS Growth
• 3-Year EPS Growth
• Dividend Yield
• Change in FY1 Estimates over 3 Months
• Up vs. Down EPS Est. Revisions
• LTM EPS Yield
• Estimated FY1 EPS Yield
• Scored Strategy Returns
• Subjective Estimates
• Optimized Estimates
• Summary

3
Establishing Long-Short Trading Strategy
Introduction

• Objective
• Generate positive returns
• Limit risk through hedging
• Quantitative stock screen
• Seven factors
• Find predictive powers on positive and negative
  returns
• Select factors with strong predictive powers
• Go long stocks in top quintile
• Go short stocks in bottom quintile

4
Description of parameters used for screening
process
Methodology

• Sample
• US equities listed on both NYSE and NASDAQ
• Market capitalization above 100 million
• Monthly data
• In-sample time frame 1988 1998
• Out-of-sample time frame 1999 2003
• Selected variables believed to best predict
  future stock returns
• Allocated factors into quintiles based on
  selected criteria
• Resampled factors each month
• Analyzed output and performance over time

5
Description of Factors
Factors

• 1-Year EPS Growth expected growth in EPS over 1
  year
• 3-Year EPS Growth expected average yearly growth
  in EPS over 3 years
• Dividend Yield indicated dividends / current
  price
• Change in FY1 Est. over 3 Months change in
  earnings estimates over a 3 month period
  (momentum play)
• Up vs. Down EPS Est. Revisions ( of Up - of
  Down revisions)/Total Estimates (momentum play)
• LTM EPS Yield yield on EPS over the last twelve
  months (EPS yield is inverted P/E ratio)
• Estimated FY1 EPS Yield FY1 EPS estimate /
  current price

6
1-Year EPS Growth not a suitable factor for a
long-short strategy
Factor 1-Year EPS Growth

• Difference between quintile 1 and quintile 5 not
  large enough
• Quintile 1 not consistently enough best
  performing portfolio, and quintile 5 not
  consistently enough worst performing portfolio

7
3-Year EPS Growth not a suitable factor for a
long-short strategy
Factor 3-Year EPS Growth

• Magnitude of returns too small and difference
  between quintile 1 and quintile 5 not large
  enough
• Quintile 1 not consistently enough best
  performing portfolio, and quintile 5 not
  consistently enough worst performing portfolio

8
Dividend Yield displays some positive predictive
ability
Factor Dividend Yield

• Quintile 5 outperforms quintiles 3 and 4 on
  average, mitigating the short portion of the
  strategy
• Quintile 1 does outperform all other quintiles on
  a reasonably consistent basis, pointing to some
  predictive power
• This factor could be used in a multivariate
  scored long-short strategy

9
Change in FY1 Est. over 3 Months has the
potential to make a contribution in multivariate
model, but not on its own
Factor Change in FY1 Est. over 3 Months

• Turnover rate is rather high which would lead to
  high transaction costs
• Quintile 1 is fairly consistent yielding the
  highest return, and quintile 5 is also fairly
  consistent in yielding the lowest return

10
Up vs. Down EPS Est. Revisions would not
guarantee returns high enough on its own, but
could be used in a multivariate model
Factor Up vs. Down EPS Est. Revisions

• Difference in returns between quintile 1 and
  quintile 5 not high enough to make this strategy
  attractive for a long-short strategy
• Factor performs very well in three turbulent
  years, 2000 2002, suggesting that it could play
  a valuable role in a multivariate model

11
LTM EPS Yield does a remarkable job in adequately
repeating the highest return yielding portfolio
Factor LTM EPS Yield

• Wide spread between quintile 1 and quintile 5
  which would make this strategy attractive from a
  return perspective
• Quintiles 1 and 5 perform as expected over time,
  except for 1999, which would have been disastrous
  and led to a return of (68)

12
LTM EPS Yield does a fairly consistent job of
outperforming the market
Factor LTM EPS Yield

• On most observations, the factor outperforms the
  market, especially during years where the market
  went down
• However, in 1999, following a trading strategy
  based on this factor would have been disastrous

13
Estimated FY1 EPS Yield is most promising factor,
with consistently high and low returns for
quintiles 1 and 5 respectively
Factor Estimated FY1 EPS Yield

• Long-short strategy generates significant
  positive return in all years except 1999 with a
  loss of (54.71)
• Including this loss, this strategy would still
  generate a 728 cumulative gain over the past 5
  years
• Consider utilizing other variables in a scored
  strategy to mitigate 1999 returns

14
Estimated FY1 EPS Yield shows significant upside
Factor Estimated FY1 EPS Yield

• In most years, a long-short strategy based on
  this factor would outperform the SP 500 Index,
  with returns exaggerated in down markets
• As discussed before, 1999 would have produced
  catastrophic negative returns

15
Considerations
Scored Strategy Returns

• Rationale
• Some factors were useful predictors of either
  upside or downside returns
• Scoring system utilizes predictive power of
  numerous factors
• 2 Methodologies
• Subjective Scoring
• Looking at historical results, determine most
  useful factors
• Assign weights using intuition and group
  discussion
• Optimized Scoring
• Construct correlation matrix of several factors
• Conduct mean-variance analysis, using data
  derived from one-factor models
• Apply optimal weights to several factors

16
Determining scored factors and weights
Scored Strategy Returns Subjective Estimates

• Subjective Estimates
• Evaluated 7 factors, but selected only 3 factors
• FY1 EPS Yield
• High correlation with LTM EPS Yield but better
  results
• (5 if 1, 1 if 2, -3 if 5)
• Dividend Yield
• Positive Performance Predictive Ability
• (2 if 1)
• Up vs. Down EPS Est. Revisions
• High correlation with Change in FY1 Est. over 3
  Months but lower turnover
• (3 if 1)

17
Very powerful predictive ability of high and low
returns
Scored Strategy Returns Subjective Estimates

• Continues to generate positive returns in each
  year except 1999, but losses are reduced to
  (25.23)
• Including this loss, this strategy would generate
  a 437 cumulative gain over the past 5 years
• Standard deviations and betas of quintiles 1 and
  5 are almost identical

18
Subjective Scored Estimates display similar trend
as previous best model (FY1 Yield), but less
volatility
Scored Strategy Returns Subjective Estimates

• Graph below depicts equal weighted annual returns
  of long-short strategy of Subjectively Scored
  Strategy and FY1 Yield Strategy
• While Subjective Scored Strategy sacrifices some
  upside, it performs much better during market
  anomaly of 1999

19
Determining scored factors and weights
Scored Strategy Returns Optimized Scoring

• Optimized Estimates
• Utilized a mean-variance optimizer
• Each selected quintile is essentially a portfolio
  with a mean and variance
• Evaluated all 7 factors in optimization model,
  and selected 4 factors (6 total quintiles)
• Estimated FY1 EPS Yield
• (3.42 if 1, 0.64 if 2, -1.20 if 5)
• FY1 Revision Ratio
• (0.04 if 1)
• Dividend Yield
• (0.92 if 1)
• LTM EPS Yield
• (-2.83 if 1)

20
Value Weighted Portfolio shows intriguing results
Scored Strategy Returns Optimized Scoring

• The equal weighted portfolio using optimized
  scoring produces very noisy results
• However, the value weighted portfolio possesses
  the favorable step distribution
• Over the past 5 years, a long-short strategy with
  the value weighted portfolio would have garnered
  a cumulative 174 gain.

21
Value weighted optimization appears to perform
slightly worst than equal weighted subjective
portfolio
Scored Strategy Returns Optimized Scoring

• The value weighted optimized portfolio produces a
  negative return twice and performs worst than the
  subjective portfolio in 1999
• The significant turnover of quintile 5 (37)
  could also pose a problem with respect to trading
  costs

22
Initial findings have 5 MBA students pondering
quitting school, rejecting their job offers, and
starting a hedge fund
Summary

• Estimated FY1 EPS Yield
• Empirically and logically a very strong factor
• In most markets and at most times, earnings
  continue to drive stock prices
• However, market anomalies such as 1999 make this
  strategy vulnerable
• Combining this factor with others should reduce
  volatility
• Subjective Scoring
• Adding reasonably uncorrelated factors drives
  down standard deviation
• Utilizing intuitive weights for variables proves
  to be a valuable exercise
• Best results of tested strategies
• We realize this is not an exhaustive list of
  long-short strategies, but are confident this
  model can produce significant returns
• A small (or large) hedge fund cannot incur losses
  of 50 or more in 1 year, so we are pleased with
  reduced volatility at expense of some upside that
  the scoring system brings