Factor Timing and Crowding

Factor Crowding Analysis

"The Impact of Crowding in Systematic ARP Investing"

Goal

Impact of factor crowding for ARP strategies profitabilities and risks

Assumptions

  • the definition of "crowdness"
    • should be the impact on return (response variable) rather than overall ARP size
      • return crash (factor crash)
      • market capacity and market impact
    • relevant research
      • equity momentum crash
      • benefit of volatility-targeting overlay
      • valuation spreads as factor timing indicators
  • two types of risk premia
    • risk premia/negative feedback loops
      • (share risk, eg. value(distressed risk), volatility) - compensation for undiversifiable risk
    • investor trading shrinks the valuation spread
      • price anomalies/positive feedback loops, momentum, behavior driven, limits to arbitrage)
      • investor trading inflates the trading signal
      • unstanstanable - eg. funding liquidity and momentum strategy
    • convergence vs divergence
      • convergence: flow in will lead negative feedback strategies to lose alpha
      • divergence flow will lead positive feedback strategies to
  • measure of crowdness
    • pairwise correlation of factor-adjusted returns of assets in the same peer group (high/low factor score)
    • motivated by Cahan and Luo(2013), Lou and Polk(2014) and Huang Lou and Polk(2018)

Findings

  • divergence strategies exhibits lower return and higher volatility after crowding
  • convergence strategies exhibits higher return and lower volatility in crowded periods

Tests

  • two types of risk premia
    • turn over of global equity momentum/value return vs 1y turnover
      • momentum negatively correlated, value positively correlated
  • measure crowdness
    • input to the system: flows data and positioning data
    • output to the system: price, movement and asset co-movement and valuation spreads
      • positioning data has problem of depth, timeliness of availability and historical and asset coverage problem
      • synchronous asset comovement (as a portfolio) might be affected by the inflow of factors
        • Cohan and Luo(2013) are the first to use pairwise correlation of a group of U.S. stocks that share a common characteristics (by factor scores, eg. top momentum) to indicate factor crowdness
    • measure:
      • peer group correlations: use a regression to get factor return, then rolling 52 week co-movement metrics
      • utilization by securities lending data: for top and bottom tier baskets (larger utilization of un-attractive basket of stocks, more crowded
  • Data
    • single stock universe
    • cross section of commodities and FX
      • 26 currency pairs against USD, 24 GSCI commodity indices
    • use ARP Allocation mechanism to form factor portfolios
    • return data
      • adjust factor portfolio return by other factors (eg. in equity, if use a 4 factor model, adjust momentum by market, value and size)
  • Hypothesis (Oct2005-May2018)
    • two samples: form a metric (co-movement) during periods of higher and lower crowding (top 20% vs bottom 20%)
    • Impact on return:
      • H-0: no impact on subsequent 2y buy-and-hold return
      • Newey and West standard errors
      • 6month period to observe most significant return effect for divergence strategies
      • H1 holds
    • Impact on risk
      • H0: after crowded period: convergence/divergence period t 1y(t-t+1y) volatility and skewness is not change
      • divergence: strong evidence more volatile after crowding period
      • convergence: weaker results crowding will stablize the portfolio
      • no statistically strong results on skewness
      • Wilcoxon Rank-Sum effect
    • volatility during transition period
      • test $$\frac{\sigma{t,t+\Delta t}}{\sigma{t',t'+\Delta t}}$$ vs next 66-day same ratios
      • strong results for convergence strategies

backtesting

  • cometric on alternative risk premia strategies

Factor Timing Methods Review

  • Background
    • efficient market does not mean zero predictability - beta factors in equities, bonds, credit, FX, real estate
    • most ARP suffer from overfitting, but shoes some predicatibility
      • in equity "anomalies", FX Carry, bond, etc
    • business cycle significantly affects value/momentum/carry strategies
  • Timing based on economic conditions
    • "nowcasting"
      • developed market vs emerging market, inflation(imported, input, wage...), market stress (liquidity, spreads, volatility...)
      • diffusion index + nowcaster
    • business cycle
      • expansion: Momentum, slowdown: min vol quality, contraction: min vol, quality, recovery: size, value

Factor Timing Effectiveness

Test the statistical possibility of factor timing

  • Assumptions: folded normal distribution and p% of time successful timing on return
  • Tests
    • Single Strategy
      • normal distribution test
      • t distribution Monte Carlo Simulation
      • test common timing signals hit ratios:
        • commodities, equities, FX, Rates/Credit with hit ratios
    • Multiple Portfolio Test
      • weighting scheme for the active weights, should increase with forecasting skill vs equal weight vs worst(all wrong)
      • active weight 20%, constant correlation simulation and block bootstrap (1000 iternations)
      • parameters (correlation (higher the better)) active weight (there is an optimal)
  • Conclusion:
    • required forecasting skill to outperform (mean return) increases substantially with high Sharpe Strategies
    • diversification (low correlation) raises the bar of timing benefit
    • next step: focus on the magnitude and strategy turnover (transaction cost)

Factor Timing Metrics

  • Used Embedded metrics as factor timing indicators
    • valuation/carry/momentum: inner product between weight vector and vector of characteristics
    • co-movement/crowding: average market-adjusted cross-correlation of strategy constituents
    • factor momentum: serial correlation over pre-determined window
  • Assumptions - works like technical analysis on factors
    • risk based characteristics: like value and carry, expected to exhibit longer-term predictive ability
    • price-based characteristics, like trend/factor momentum, expected to exhibit short-term predictive ability
    • crowding; expected to have medium term predictive ability
    • relies on mean reversion
  • Analysis
    • in-sample analysis: regression with future strategy performance vs . average monthly metric over multiple horizons
      • linear
      • non-linear: convexity on left tail- rebound, concavity on right tail - crash
    • out-of-sample analysis: timed strategy use the metrics
      • tilt based on capped z-score
  • Caveats
    • high-turnover strategies: may not able to use long-term signal
    • hindsight bias and artificial correlation of longer horizons

ESG as a Factor

  • S
    • ESG Could be related to productivity (eg. work-life balance), Use ESG as a single factor or as a filter (filter out low scoring ones - overlay)
  • T
    • test the effectiveness as a factor
    • control ESG risk will filter out small caps
  • A
    • try ESG as a factor score model and long-short portfolio
    • test correlation with existing factors
    • make part of another factor
      • eg. Quality + ESG
    • as a filter (parameter: thredhold)
  • R
    • no statistical effectiveness as a factor
    • with filtering, can achieve similar risk/return with better ESG Score
  • Potential Problems/Challenges
    • no standard/transparent ESG Score
    • data is limited (new thing)

Risk-Premia Factor in Strategic Asset Allocation

  • Problem: Substitute traditional assets with lower cost and more robustness
  • Assumptions: based on PCA (or KNN) analysis to cluster risk-return characteristics
    • Eq like: Volatility Carry(short Vol) Strategies
    • Duration like: low beta equity, unconstrained bond strategies, etc
    • CTA like: momentum and time series momentum strategies
    • Diversified: Value, Size, Curve
    • Dispersion and Structure imbalance like strategies
  • Analysis: Comparing with standard (risk parity or generalized risk parity) cross-asset portfolio, with different constraints( eg. beta control, drawdown control)
    • substitute similiar strategies
    • compare conditional Sharpe (or similar) performance/drawdown

results matching ""

    No results matching ""