Market Cycle Varying Multifactor Strategies

1
Market Cycle Varying Multifactor Strategies

• Cyclical Analysts
• Noah Harris
• Juliet Xu
• JJ Haines

2
Goal

• To determine whether or not different factors
  behave better or worse during different stages of
  the economic cycle.
• To develop a time-varying multifactor strategy to
  take advantage of these differences in factor
  behavior.

3
Agenda

• Overview
• Factors
• Market Cycles
• Scoring Screens
• Results
• Conclusions

4
Overview

• Identify successful factors from Global Asset
  Allocation final projects
• Run in- and out-of-sample screens for all factors
  (monthly)
• Use historical data from a pre-determined
  four-stage cycle to segment factor returns
• Regress macroeconomic data to predict next
  months cycle stage

5
Factors

• We looked at the best factors used by the groups
  from Global Asset Allocation.
• Selected Factors
• Book to Price
• Analyst Estimate Revision
• Price Momentum
• Free Cash Flow Yield (FCF/P)
• NTM Forward Earnings to Price
• Fundamental Debt Factor eliminated due to poor
  in-sample performance

6
Market Cycles

• Defined by four liquidity regimes
• Calm positive increasing
• Speculation positive decreasing
• Turbulence/Crisis negative decreasing
• Rebound/Recovery negative increasing
• (where positive/ negative refers to the liquidity
  environment)

7
Factor In-Sample Performance
8
Scoring Screens and Factor Weightings for
Different Cycles
Factors are scored from -5 to 5 by each teammate
and averaged Long Weightings sum to 100, Short
Weightings sum to -100 Total represents
weights for a non-varying multifactor scoring
screen
9
Strategy Performance-Annual
The stable screen had the best out-of-sample
performance for both long only and long-short
but the cycle varying screen beat the S P more
often in long only.
Cycle Varying- Scoring screens vary with market
cycle Equal Wt Factors Varying- Scoring screens
vary with market cycle factors are equal
weighted Stable-Scoring screen stays constant
through all periods
10
Overall Performance(Out-of-Sample)
11
Strategy Performance-Cycles
Both the Cycle Varying and Stable Long Only
strategies beat the S P in all four market
cycles. The Cycle Varying Long-Short beat the S
P in 3 out of 4 cycles.
12
Mistiming

• QuestionWhat happens if we miss the turning
  point from one regime and the next?
• Check returns out-of-sample assuming that we miss
  the turning point by 1, 2 or 3 months
• The results show that there is little or no
  significant loss in Alpha!

13
Prediction for Next Period

• PredictionNext Period will be CALM
• Procedure
• Use Logistic Regressions for each regime
  independently and compare results

14
Example of Predictive Process

• Use all monthly data for final prediction
• Use backward stepwise to optimize the number of
  variables (P-to-remove 0.05)
• Results in different final list of variables for
  each regime

ExampleTest for Calm
15
Conclusions

• Using the cycle varying strategy adds positive
  alpha
• The stable investment strategy outperforms the
  cycle varying strategy out-of-sample
• While both our stable and cycle varying
  strategies performed well, the stable strategy
  performed better for long-short and long only and
  had higher alphas
• Our cycle varying strategy had more consistent
  performance, beating the market in 7 out of 8
  years, but was not clearly able to add alpha or
  returns over a stable strategy
• Best Strategy (highest alpha) Buy the long-short
  stable investment portfolio
• Regimes are based on regressions, and therefore
  highly predictable
• Missing a regime change by up to 2 months is does
  not seriously impact portfolio performance

16
Recommendations for Further Study

• Determine which factors specifically failed out
  of sample
• Select less correlated factors
• Try to identify better performing regime
  definitions of the economic cycle
• Determine whether varying weights and factors in
  different regimes adds significant costs
• Does it cost more to vary the weights or the
  factors?