Axioma Overview
Transcription
Axioma Overview
“I want a second opinion” Understanding and Leveraging Statistical Risk Models in Portfolio Management June 14th, 2016 Chris Canova, CFA – Vice President, Head of Equity Solutions Anthony Renshaw, PhD – Equity Solutions Specialist Chris Martin, MFE, CAIA, CIPM – Equity Solutions Specialist Agenda • Axioma Risk Models and Use Cases • Types of Risk Models • Explaining Statistical Models • Case Study – US Large Cap Core Manager • Framework for using Statistical Models • Recent Trends and Observations • Improved Risk Management and Portfolio Analysis Better Results! • Q&A 2 Axioma US4 Equity Risk Model Suite • Premise: Deliver multiple risk models on a daily basis to provide the most comprehensive analysis along with the most choice and flexibility for portfolio construction. • Fundamental models with comprehensive factors for intuitive and detailed risk analysis and performance attribution • Statistical models can provide important additional risk and portfolio construction insights • Short and Medium Horizon Model variations to increase risk awareness in fast changing environments, tuned to the portfolio strategy horizon • Macro factor analysis to understand sensitivities for analyzing and stress testing the impact of macro events • Innovative methods - Robust Statistics and Dynamic Volatility Adjustment (DVA) • Model Customization - The ability to customize important model features such as the factors, horizon, and estimation universe to get the most out • Extensive factor libraries from Axioma + the ability to incorporate your own libraries or those through third-party Axioma partners such as CS HOLT • A deep daily history with historical coverage since 1982 for most model variations • Integration with the Goldman Sachs Shortfall model 3 Benefits of Axioma’s Multiple Risk Models • Confirm multiple estimates of risk Side-by-Side – “The whole is greater than the sum of its parts” • Investigate causes of differences in estimates as they occur – leading to improved knowledge of the risk environment and potential issues • Higher confidence in portfolio risk interpretations = better portfolio management • Trends distinguish differences in short term risks vs. persistent factor changes • Using Fundamental, Statistical, and Macro models from different vendors can be problematic: • • • • Different frequencies used for factor and specific returns Different methods for estimating volatilities and specific risk Different half-lives of the models Comparison analysis can be misleading 4 Understanding Statistical Models • Like fundamental models, the following returns model is used to define the factor returns and model structure: • R = Bf + u • B represents the model factor exposures • f represents the model factor returns • u represents the specific return • In a fundamental risk model, B is pre-calculated, f is estimated • In a macroeconomic model, f is pre-calculated, B is estimated • In a statistical risk model – both B and f are estimated simultaneously in order to maximize model explanatory power • Benefit – statistical models are “free” to find sources of risk that may be different from those pre-imposed by the model structure 5 Overview of Model Methodology Benefits • Fundamental Models • Assumes factors are known in advance, estimates factor relationships • Widely used for client reporting and communication • Consistent framework for risk decomposition, performance attribution and portfolio construction – can explain the factors driving current risk exposures in line with factors explaining historical performance • May not contain “all factors” relevant to practitioners • May miss short-term temporary factors • Statistical Models • Makes no assumption of the factors in the model, allows factors to emerge based on the estimation process (simultaneously identifies factor exposures and factor returns) • Structure provides the ability to capture short-term risk phenomena • Responsive and performs well in portfolio construction • However, lacks intuition of factor exposures when used in isolation • Macro Economic Risk Models • Provides sensitivities to key macro factors • Provides a stress testing framework to understand portfolio responses to macro events/surprises 6 Typical Risk Analytics Best Practices • Risk decomposition to isolate the sources of expected risk • % contribution to risk coming from common factors vs. specific risk • Point-in-time (most recent) with historical time series trends • Track factor exposures to ensure portfolio consistency • Understand trends in style, industry, country, and currency exposures through time • Asset-level risk analysis • Understand the concentration of risk at the asset or grouped asset level • % of risk, marginal contributions, implied expected return • Factor-based performance attribution • In the context of the risk model – are portfolio exposures paying off on a riskadjusted basis? • Understand sources of return as a feedback loop to portfolio construction • Reduce or remove problematic exposures and ensure asset-level return expectations are translating into ongoing portfolio returns 7 What is missing? Can we do better? • A common goal in any investment process is to: • Ensure the portfolio is taking exposures in line with the investment process • Assets, factors, and investment themes • Identify, understand, and potentially eliminate undesirable risks • What if the fundamental model is missing important factors? • There may be a grouping of assets that ex-post have surprising factor returns • Using a single model could expose the portfolio (and often times lead to) problematic exposures… • How can we identify this additional risk? • Our case study will propose additions to the typical process 8 Case Study – US Active Manager • US Large Cap Core Portfolio • Typically holds 50-100 stocks (Quarterly holdings frequency) • Benchmark: Russell 1000 • Target: 3.5% - 4% annualized tracking error • Seeks to provide exposure to momentum and valuation-oriented factors for long-term active out-performance • Optimized using asset rankings while seeking to control active risk within the tolerance range • Includes sector and other style constraints • Targets 80%-100% annualized turnover 9 Active Risk Trend – Fundamental Model 10 Active Risk Trend – Multiple Models 11 Zoom to Last Nine Months 12 Factor Risk Spreads • Definition: • Factor Risk Spread – The highest minus the lowest factor risk estimate • Implications: • Allows the client to monitor the general level of risk differences across risk models to identify important trends and when differences arise in systematic risk • In Practice: • Time series analysis with daily updates provides early signals of possible changes in risk regime and/or the emergence of non-traditional factor risk sources 13 Statistical versus Fundamental Risk Spread • Definition: • Statistical risk model estimate minus the fundamental risk model estimate • Implications: • The statistical risk model identifies factor volatility that the fundamental risk model has quantified as specific risk, or may have missed altogether. The risk is coming from a factor that is not included in the “traditional” factor set. • In Practice: • In mid-2008, statistical model risk estimates expanded greatly and the fundamental model risk estimate lagged. This signaled an early increase in volatility as well as different risk sources. • In mid/late 2015 and at the end of Q1 2016, we observed an additional shock in the higher estimate of risk from the stat models compared with the fundamental models. 14 All Models – Risk Trends and Spreads 1.50 1.00 0.50 0.00 SH Statistical - MH Fundamental Spread SH Statistical - SH Fundamental Spread 15 Dig a bit deeper – March 31, 2016 16 Intuition via the Assets • % of active risk is a useful analytic – We can decompose the contributions to risk coming from each asset in the portfolio and benchmark • All assets in the portfolio and benchmark will sum to 100% • Allows for useful groupings 17 So where do we go? • We can compare the % of Active Risk for each asset in the context of multiple risk models to gather some intuition as to where the differences in risk are coming from. • We now have information that can help us make some decisions around our comfort level with these increased risks and where they come from. 18 What Does History Tell Us? PROCEDURE #1 – Goal: See how risk differs within groupings • Take assets in the Russell 1000 universe • Partition into deciles by Asset Risk difference (Stat minus Fund) • Review comparisons of predicted and realized risk organized by year and deciles PROCEDURE #2 – Goal: See how return differs within groupings • • • • Take assets in the Portfolio and Russell 1000 universe Partition into quintiles by difference in “% of Active Risk” (Stat minus Fund) Create portfolios of Q1, Q5, and Q2-4. Compute performance statistics of the resulting portfolios 19 Risk Differences Identify Riskier Assets 2016 exhibited some of the highest realized risks over the last 8 years. 20 Adjusting Positions – Performance Impact Long term, HI %∆ TE hurt performance; since 2014, it was a winner. 21 How can I use this information? • Quantitative Active Managers • Introduce a second risk constraint or objective penalizing risk coming from the statistical model (in general, or when spreads suggest necessary) • Adjust asset-level constraints to reduce exposure to assets with high delta • Pre-screen for risk differences and incorporate into portfolio construction • Low Vol – clear pattern of higher realized vol in these assets – Low Vol strategies – research opportunity • Fundamental/Quantamental Active Managers • Adjust position sizes for problematic assets to ensure conviction is properly implemented • Long-Short Managers • Explicitly hedge systematic risk as estimated by the stat models in addition to fundamental model • You are not factor neutral if you are optimizing with fundamental models • There is a better “best hedge” • Constrain assets with increased risk coming from the stat models • Early warning signal on potential problem areas • Passive/ETF/Tax Efficient Managers • Constraint tracking error using multiple risk models • Tighten asset bounds for assets with larger differences in risk estimates 22 Recap of Recommended Approach Step 1 - Risk Spreads • Compare multiple risk predictions on the same graph/at the same time • Extract current and historical maximum risk spread • Examine different time periods – recent vs. long term Step 2 - Risk Decomposition/Projection • Fundamental risk model: % factor risk, % specific risk • Statistical risk model: % factor risk, % specific risk • Leave out the decomposition of risk at the factor level • Projected statistical risk model: what % of risk is unexplained by fundamental factors Step 3 - Asset Level Analysis • Rank positions by “% of Risk” or “% of Active Risk” or their differences • Outliers account for a large fraction of risk budget – are those bets intentional/worth it? Step 4 – Seek to identify the source of this factors • Custom Risk Model Using “Stat minus Fund” Factor – quantify return contribution from this exposure • Can be used to indicate which standard fundamental factors are most impacted by the missing factor risk • Identify other factor types that could be used to overlay into the Projected Risk Decomposition (Alpha Factors, Macro Factors, Trading/Crowding Factors, etc.) 23 Summary • Fundamental models have articulated and captured the return swings and risk drivers to important factors such as value, momentum, and volatility over the recent past • However, there is additional systematic risk/return that needed explaining…. • Over this period we have seen (two separate instances) that additional systematic volatility has existed in the market that the statistical models are properly capturing • We articulate a framework to identify those risks, quantify the impact, and establish intuition into the sources of those risks in a meaningful and actionable framework • The effort and cost required to implement the proposed framework is low using Axioma • Any portfolio manager, quant, or risk specialist tasked with managing exposures and risks in order to maximize risk-adjusted returns should use the proposed framework in an automated ongoing process • Reach out to your Axioma consultant to learn more! 24 Axioma Integration on FactSet • Portfolio Construction • Axioma Portfolio Optimizer • Portfolio Simulation • Portfolio Batch Optimizer Insight article: “Moving From Research to Construction” Part 1: Stock Selection Part 2: Constructing Portfolios Part 3: Simulation Performance 25 Axioma Integration on FactSet • Performance and Risk Reporting • Portfolio Analysis • Portfolio Dashboard • Stress Testing 26 Thank You For additional information, contact: [email protected] Visit: www.axioma.com