Portfolio Optimization. OMAM Quantitative Strategies Group. OMAM at a glance. Equity investing styles: discretionary/systematic spectrum

Size: px

Start display at page:

Download "Portfolio Optimization. OMAM Quantitative Strategies Group. OMAM at a glance. Equity investing styles: discretionary/systematic spectrum"

Clarence Chase
5 years ago
Views:

1 CF963, Autumn Term Learning and Computational Intelligence in Economics and Finance Part 1: Introduction to quantitative investing in hedge funds Part 2: The problem of portfolio optimisation Part 3: Application of heuristic portfolio optimization Portfolio Optimization University of Essex 2 nd December 2013 Dr Amadeo Alentorn Head of Quantitative Research amadeo.alentorn@omam.co.uk OMAM at a glance OMAM Quantitative Strategies Group > Dynamic investment firm, focused on high performance and absolute returns > Independent investment teams no house view or investment style > Expertise in Discretionary equities Academic Advisory Board Investment research Fixed income Quantitative strategies > Range of long only and alternative products designed to meet the needs of retail and institutional investors Investment team > New product innovation > Independent risk management team > USD6.5 billion assets under management* > Part of Old Mutual Group, a constituent of the FTSE 100 index. V-Lab: PhD and post-doctoral researchers Systems development Portfolio construction and management * Source: OMAM, as at 30/10/ Equity investing styles: discretionary/systematic spectrum Types of equity funds: passive/active spectrum > Discretionary investing Mostly driven by fundamental analysis Analysis of company s accounts, business plans, competitors, etc Meeting management and understanding business model Usually focus in one country/ sector/ industry Longer holding horizon > Quantitative investing Factor models based on both fundamental and technical analysis Able to analyse and evaluate 1000s of companies Highly diversified portfolios Medium term holding horizon > Statistical arbitrage Usually based on technical analysis, price driven (i.e. pairs trading) High frequency execution, usually intra-day trading Short holding horizon Discretionary Systematic > Across the industry, many different variations: quantamentals, quant overlays, etc. > Index tracking Tracking error TE < 0.25% > Enhanced long only TE between 0.50 to 1.50% > Active long only TE greater than 2.00% > Short extension (130/30) Equivalent to index plus 30/30 market neutral > Market neutral hedge funds Dollar neutral, i.e. 100/100 > Long/short hedge funds Variable net exposure Long only funds Hedge funds > Trade-off between risk and expected return. Passive Active 5 6 1

Types of equity funds: long, short and net positions > Given $100 of capital, the structure of the portfolio will be very different, for different types of funds.

2 Types of equity funds: long, short and net positions > Given $100 of capital, the structure of the portfolio will be very different, for different types of funds Long only Short extension Market neutral Long short 7 Long book Short book Net position Performance of a market neutral fund vs. market > Market risk is the primary source of risk for a long only equity fund in absolute terms When equity markets fall by 30%, the best outcome for a LO fund is -30% + TE. Market neutral designed to deliver positive returns in all market environments 10 0 Jan94 Sep95 May97 Dec98 Aug00 Apr02 Nov03 Jul05 Feb07 Oct08 Return Risk Sharpe Sortino MSCI World Index Market neutral hedge fund Short-selling Short-selling: Northern Rock case study > Traditionally investing (long only constraint) can create miss-pricings, because only positive views about stocks can be fully reflected into prices Northern Rock Plc > Hedge funds are able to reflect buy and sell ideas into the portfolios, by buying stocks with positive forecasts, and shorting stocks with negative forecasts > How does it work? Broker/dealers on behalf of clients identify stock owners (i.e. pension funds) Hedge fund borrows stock, and sells it on the open market At a future date, hedge fund buys back the stock, and returns it to the lender Stock owner receives a lending fee for the service Note: naked short selling is banned! > Risks Stock recall Short squeezes: when prices go up, rush to cover positions Potential of infinite loss Jun07 Aug07 Oct07 Dec07 Feb08 Apr08 Jun08 Aug08 Oct08 Dec08 Feb09 Utilisation Apr09 Jun09 Price Aug09 Oct09 Dec09 Feb10 Apr10 Jun10 9 Source: Data Explorers 10 Types of equity hedge funds Our investment approach and philosophy > Hedge fund indices have become a useful tool for monitoring the performance of the different hedge fund strategies But issues with survivorship bias, self-selection bias (voluntary reporting), etc > Useful indication of risk associated with each strategy > Market inefficiencies result from behavioural biases and structural anomalies > Investment insights, not statistics, drive research Many of the investment strategies are not different than those used by fundamental fund managers (i.e. valuation, earnings quality, etc) > Rigorous backtesting of data to validate investment criteria One of the key advantages over fundamental managers, we are able to test the historic performance of an investment strategy > Continuous research to maintain and enhance information ratio To mitigate risk of imitation and of crowded trades > Multi factor models designed to perform throughout the business cycle Factor modelling lends itself to market neutral investing, generating both buys and sells Source: Hedge Fund Research

understanding of how human behaviour influence prices, and helps to explain some of the observed miss-pricings and anomalies: Overconfidence: all traders are above average, over-trading Prospect

3 Investment strategies based on behavioural finance Academic research vs. industry research > Financial markets are complex systems Prices are the result of complex interactions between market participants > Academic literature on behavioural finance aims to provide an understanding of how human behaviour influence prices, and helps to explain some of the observed miss-pricings and anomalies: Overconfidence: all traders are above average, over-trading Prospect theory: people value gains and losses differently Anchoring Under-reaction/ overreaction / herding Naïve extrapolation > A solid understanding of the rationale for an observed market anomaly from a behavioural finance point of view, when researching investment strategies, helps to avoid falling into data mining traps. > Quantitative investment allows to remove emotion from the investment process 13 > Liquidity Many academic studies use the CRSP database as the universe of stocks on which to test a hypothesis, up to 26,000 stocks. In practice, only around 3,500 stocks worldwide with enough liquidity to invest with negligible market impact. > Survivorship bias Important to include companies that are no longer in the universe (impact of bankruptcies/corporate actions) > Reporting lags Point in time databases. > Backfilling See Rewriting history > Transaction costs Specially for high frequency strategies 14 Factors: academic debate alpha vs. risk? (Fama French) Quantitative investment process 10 2 Market Rtrn Rsk IR Sortino Size Value RF 3.6 > Fully systematic investment process Universe: top 95% of market capitalisation in each country: around 3,500 stocks 1 Multi factor stock selection criteria Stocks ranked by attractiveness Optimise risk and return profile of portfolio Final portfolio Source: Kenneth French website Portfolio optimisation process Part 1: Introduction to quantitative investing in hedge funds Part 2: The problem of portfolio optimisation Trade cost model Part 3: Application of heuristic portfolio optimization Relative return model Optimisation and portfolio construction Risk model Continuous research on all three proprietary models and process Trade list 17 3

4 What s portfolio optimisation? The mean-variance frontier > Given a universe of stocks, and a set of expectations about the future performance of these stocks, the problem is how to construct a portfolio by selecting how to allocate the capital. > The classic model for portfolio construction is the mean-variance optimisation introduced by Markowitz in > A portfolio optimization problem typically consists in maximising return, minimising risk or maximising utility by finding the optimal set of stock weights (i.e., percentages of invested capital) that satisfies a set of constraints The traditional optimisation problem Long vs short stock positions > The traditional portfolio optimisation problem with no shorting allows only positive (long) stock positions: > Long: a long position means the investor has gone to the market and bought some shares in a company. The proportion of the value of these shares of the total portfolio is the portfolio weight of that company in the investor s portfolio, represented by a positive portfolio weight. > The mean-variance objective function represents a trade-off between expected return and expected risk, weighted by a risk aversion parameter λ: > Short: a short position is achieved when an investor sells shares that does not own, and this is represented by a negative portfolio weight The optimisation problem for a market neutral hedge fund Why heuristic optimisation? > For a market neutral hedge fund, we remove the no-shorting constraint to allow negative weights: > Stock returns exhibit non-normal characteristics, including skewness and fat tails, and most investors are loss averse > These invalidate the assumptions in the traditional approach. > Heuristic optimisation methods provide a more flexible toolset where no simplifying assumptions are needed. > Some of the applications have successfully achieved: > Objective functions: > Mean Variance > The use of non-quadratic risk measures such as Value at Risk > Incorporation of integer constraints, such as cardinality constraints. > Heuristic algorithms already used to tackle this problem: Genetic Algorithms, and > Mean Value-At-Risk Threshold Accepting algorithms > In the last part of the lecture, we will look at a new heuristic algorithm: the GNMA

Part 1: Introduction to quantitative investing in hedge funds Part 2: The problem of portfolio optimisation Part 3: Application of heuristic portfolio optimization Introduction > This work is based

5 Part 1: Introduction to quantitative investing in hedge funds Part 2: The problem of portfolio optimisation Part 3: Application of heuristic portfolio optimization Introduction > This work is based on a recent paper presented at UKCI Heuristic Portfolio Optimisation for a Hedge Fund Strategy using the Geometric Nelder-Mead Algorithm > We present a framework for implementing a heuristic portfolio optimisation framework for a market neutral hedge fund investment strategy. > We also illustrates the application of the recently developed Geometric Nelder-Mead Algorithm (GNMA) in solving this real world optimization problem, and compare it with a Genetic Algorithm (GA) approach Nelder-Mead Algorithm (NMA) > Numerical optimization method published in 1965 by Nelder and Mead! > Heavily used by practitioners in optimisation > Cousin of Particle Swarm Optimization (PSO) and Differential Evolution (DE): - it does not use derivative information - it can be seen as a population based evolutionary algorithms with special operators Operations in geometric transformation > NMA applies these 4 operations until convergence: > Worst solution reflected about the centre > If R is the best, expand in this direction > If R is the worst, we may have overshoot > If nothing has worked, shrink towards best > Its search is explicitly geometric and attempts to adapt to the fitness landscape Geometric Generalization & Specification > Recently, Moraglio and Johnson (2010) generalised the NMA from continuous to combinatorial spaces. > Reflection, expansion, contraction, shrinking operations can be naturally defined in terms of distance relations among points and generalized to spaces equipped with a formal notion of distance > By plugging a new distance in their definitions, we can derive the same operators for a new space: by using the Hamming distance on binary strings, we can derive these operators for binary strings A. Moraglio and C. Johnson. Geometric generalization of the nelder mead algorithm. In Proceedings of the 10th European Conference on Evolutionary Computation in Combinatorial Optimization,

Binary Encoding of Portfolio > Fixed set of reference stocks > The composition of the portfolio is represented by a binary string > Each stock is associated with two bits: > Presence bit: stock

6 Binary Encoding of Portfolio > Fixed set of reference stocks > The composition of the portfolio is represented by a binary string > Each stock is associated with two bits: > Presence bit: stock included or not > Sign bit: long (0) or short (1) position > Quantity of stock: budget equally partitioned on shorts and longs to have a neutral portfolio Data > Daily stock returns for the 100 names of the FTSE100 index between Jan 2007 and Dec 2009 > Covariance matrix based on full covariance with a rolling window of 250 days > VaR based on the historical method, the 99% VaR is the third worst point over the last 250 days > Stock return forecasts are such that by construction reflect some level of predictive power (5%), with 95% pure noise Experiments Results: quality of solutions > Two optimisation algorithms: > Similar fitness levels achieved by both algorithms > GA vs GNMA > Similar execution time taken to achieve best fitness by both algorithms > Interested in fitness and execution time > Two objective functions (strategies): > Mean-variance vs mean-var > The new GNMA algorithm is comparable with the established GA approach. > Realised risks across strategies is comparable. > Risk aversion parameters were calibrated so that the risk levels of the two strategies were similar > Interested in differences in investment performance > A total of 4 strategies > Portfolios optimised over 250 days, average 10 runs Results: investment performance Findings of the paper > Portfolios optimised for VaR deliver higher returns, as they take advantage of the asymmetry in the risk penalty function > Better reflects investor preferences about loss aversion and downside risk > The results validate the models chosen: > Return: we have shown that a low level of information content in return forecasts (95% of noise) delivers realistic positive returns > Risk: forecasted vs. realised risks are in line > We have shown for the first time how the recently developed GNMA method is suitable for tacking a real world problem, delivering a performance in line with the GA approach. > Optimising using VaR instead of variance as the risk measure for a market neutral portfolio limits downside risk and is able to deliver higher returns, while variance limits both downside and upside potential and therefore results in lower returns

7 Conclusion of the lecture > Overview of the world of quantitative investing in practice > Issues encountered in portfolio optimization: Size of problems: hard to move away from traditional frameworks Number and types of constraints Inputs are estimates: how to incorporate errors? > Companies doing interesting work in portfolio optimisation: Market leader in portfolio optimisation software: Robust portfolio optimisation software: Heuristic methods for random portfolio analysis: Thank you! Questions? 37 7

Grant Park Multi Alternative Strategies Fund. Why Invest? Profile Since Inception. Consider your alternatives. Invest smarter.

Consider your alternatives. Invest smarter. Grant Park Multi Alternative Strategies Fund GPAIX Executive Summary November 206 Why Invest? 30 years of applied experience managing funds during multiple market