50 ways to beat the benchmark

Transcription

1 ETF Research Academy 50 ways to beat the benchmark 1 50 ways to beat the benchmark Designing optimally diversified Smart Beta ETFs This document is for the exclusive use of investors acting on their own account and categorised either as eligible counterparties or professional clients within the meaning of markets in financial instruments directive 2004/39/ce Research Summary February 2017

2 A big data approach to selecting characteristics can offer a risk-adjusted return 140% higher than that of the standard, market capitalisation-weighted portfolio 1. Introduction 3 How many company characteristics do we need? 4 The effect of trading diversification 5 Definitions of characteristics 6 The benefits of a big data approach 7 Conclusions 8 About the authors 9 About the ETF Research Academy 10 Research paper Abstract 11 1 Source: CRSP, Compustat, I/B/E/S from Jan 1980 to Dec It includes all stocks from NYSE, AMEX, and NASDAQ exchanges and 51 firm - characteristics. The final data set has 17,930 firms on which 3,071 firms have data every month. Past performance is no guide to future returns.

3 ETF Research Academy 50 ways to beat the benchmark 3 Introduction 50 ways to beat the benchmark The latest research from the ETF Research Academy*: 50 Ways to Beat the Benchmark? Designing Optimally Diversified Smart Beta ETFs looks at how to select the characteristics or factors that can lead to a better smart beta equity ETF. Academics and investors alike have long sought to understand which anomalies or characteristics can cause shares to generate above-market performance. Standard approaches have focused purely on the return characteristics of particular sets of shares, using statistical analysis to determine why some perform differently to others. The standard approach, however, is incomplete. Portfolio builders and managers are concerned not only with potential returns, but the resulting implications for risk and transaction costs. In their recent paper for the ETF Research Academy, Victor DeMiguel of London Business School, Alberto Martín-Utrera of Lancaster University Management School, Francisco Nogales of the University Carlos III de Madrid and Ramon Uppal of the Centre for European Policy Research and Edhec Business School pose three important questions: 1. Which firm-level characteristics are significant for portfolio construction and why? 2. How does the answer to this question change in the presence of transaction costs? 3. How can investors exploit a large set of characteristics to improve the estimated performance of their portfolios? The research is of particular relevance for the designers of smart beta or active ETFs, which seek to outperform the general market. *See page 10 for more information on the ETF Research Academy.

4 ETF Research Academy 50 ways to beat the benchmark 4 How many company characteristics do we need? Over recent decades, academics have identified hundreds of characteristics that may plausibly be able to explain the behaviour of share prices over time. Such characteristics (or factors) include companies size, relative valuation, momentum, profitability, and so on. In practice, not all of these theoretical factors behave independently of each other. Most academics and practitioners accept that the set of factors can be reduced to a much smaller number of characteristics for use in portfolio construction. Most existing factor indices are based on a value judgement about the validity and attractiveness of particular factors for investment. An index designer prejudges which factor(s) are worthy of selection, perhaps because of intuitive appeal, superior past returns or particular risk-return characteristics. The index designer then creates a model to measure stocks exposures to the chosen factor(s) and builds a factor index from the factor scores produced by the model. By contrast, DeMiguel, Martín-Utrera, Nogales and Uppal select factors from a starting list of 51 theoretical characteristics using a technique called Lasso, a statistical tool from the academic discipline of machine learning. Rather than prejudging which factors might be attractive, they start with an agnostic stance and select only those characteristics that have the highest explanatory power. The only choice the researchers make is how they standardise the characteristics chosen via this machine learning process. In this way, using data on US share prices from 1980 to 2014, DeMiguel, Martín-Utrera, Nogales and Uppal identify six significant characteristics that have provided improved risk-adjusted returns, before portfolio transaction costs are taken into account: Unexpected quarterly earnings Return volatility Asset growth 1-month momentum Gross profitability Beta (a short position in low-beta stocks and a long position in high-beta stocks) The first five of these characteristics, say the authors, have historically increased mean returns while also reducing risk; the sixth, beta, has had no return effect, say the authors, but it has utility as a factor since beta helps reduce the overall risk of a portfolio based on shares characteristics. Before transaction costs > 50 theoretical factors 6 factors significant in practice

5 ETF Research Academy 50 ways to beat the benchmark 5 The effect of trading diversification In their analysis, DeMiguel, Martín-Utrera, Nogales and Uppal then take real-life constraints faced by investors (like transaction costs) into account. A surprise result of the research is that including transaction costs increases, rather than decreases the number of significant equity market characteristics. Including transaction costs raises the number of significant characteristics from six to fifteen, say the authors. After transaction costs > 50 theoretical factors 15 factors significant in practice The reason for this apparently counterintuitive result is the low or negative correlations between characteristics. To understand the result, consider an equity portfolio designed to reflect a single characteristic. The portfolio requires regular rebalancing: for example, a portfolio aiming to reproduce the characteristic of one-month momentum needs to increase its weighting in high-performing shares each month to reflect the growing momentum of such shares. At the same time, it has to cut its holdings in shares whose momentum is falling. In turn, this activity generates transaction costs. Expressed in numerical terms, when the fifteen characteristics are combined, the turnover of the portfolio is reduced by around 60% on average by comparison with a portfolio of the individual characteristics traded in isolation. The implication of this finding is that an ETF (such as an active ETF or an ETF tracking a smart beta index) which combines a number of different characteristics will face reduced transaction costs and therefore perform better than an ETF which concentrates on one or a small number of equity market factors. Again, the research approach taken by DeMiguel, Martín- Utrera, Nogales and Uppal is in contrast to the standard method of constructing factor indices. In the standard method, factors are usually selected by the index designer without taking into account transaction costs (even if costs may be taken into account at a later stage in the index design, for example by arbitrarily modifying the index rules to limit turnover). However, in the approach described in this paper, transaction costs play an integral role in the selection of characteristics. In turn, this allows for a robust portfolio design in the presence of the unavoidable trading costs facing each investor. But the researchers find that the rebalancing activities associated with each of the characteristics are relatively uncorrelated, meaning that a portfolio of characteristics benefits from what the authors call trading diversification.

6 ETF Research Academy 50 ways to beat the benchmark 6 Definitions of characteristics The fifteen significant characteristics once trading costs are taken into account are the following: Characteristic The ratio of research and development (R&D) spending to market capitalisation Asset growth Unexpected quarterly earnings Share turnover Definition R&D expense divided by end of fiscal year market cap Annual % change in total assets Unexpected quarterly earnings (IBES actual earnings minus median forecasted earnings if available, otherwise seasonally differenced quarterly earnings before extraordinary items from Compustat) divided by market cap at end of most recent fiscal quarter Average monthly trading volume for the three months t-3 to t-1 scaled by number of shares outstanding at end of month t-1 Return volatility Standard deviation of daily returns in month t-1 Volatility of share turnover Zero trading days Industry-adjusted change in asset turnover Change in tax expense Beta Financial statements score Gross profitability Industry sales concentration Monthly standard deviation of daily share turnover Turnover-weighted number of zero trading days for most recent 1 month Fiscal year mean-adjusted change in sales divided by average total assets, based on 2-digit Standard Industrial Classification % change in total taxes from quarter t-4 to t Beta estimated from 3 years of weekly firm and equal-weighted market returns ending month t-1 (with at least 52 weeks of returns available) Sum of 9 indicator variables that form fundamental financial health F-score Sales minus cost of goods sold divided by lagged total assets Fiscal year sales concentration [sum of squared % of sales in industry for each company], based on 2-digit Standard Industrial Classification 1-month momentum Return in month t-1 Industry-adjusted book to market Industry-adjusted book to market

7 ETF Research Academy 50 ways to beat the benchmark 7 The benefits of a big data approach Many equity market factor approaches focus on characteristics that have long-standing acceptance in academic literature, such as size, value, momentum and profitability (or quality). Moving beyond standard equity factors to analyse a larger set of potential characteristics has notable benefits, say DeMiguel, Martín-Utrera, Nogales and Uppal. Exploiting the larger set of characteristics results in superior estimated performance over time, say the authors. Measured in terms of the resulting Sharpe Ratio 2 and after taking transaction costs into account, what the authors call a big data parametric portfolio offers a risk-adjusted return 140% higher than that of the standard, market capitalisation-weighted portfolio 1. When compared with a portfolio of the three most commonly used equity characteristics (size, momentum and value), the Sharpe ratio improvement is 100%. To check the robustness of their results, the researchers perform a number of tests, including exploiting characteristics only after their publication in academic literature, constraining portfolio leverage, limiting turnover, controlling for stock liquidity, setting short sale constraints and using different risk aversion parameters. Designers of smart beta indices and ETFs, who often work with back-tested data, also face the challenge of ensuring that their approach is credible, given the temptation to overfit models to past results and the tendency of markets and market relationships to change over time. To bring the index design approach more closely into line with the real-life experience of investors, who select stocks or bonds based on the available data (i.e., ex ante) and then test their portfolio s performance over time, DeMiguel, Martín-Utrera, Nogales and Uppal use the past 100 months of data to select characteristics. They then keep the choice of characteristics for one year before re-estimating the model, based upon the new historical period of 100 months. In the charts below, they illustrate the hypothetical historical performance of the big data parametric portfolio (significant characteristics portfolio) without and with transaction costs. For the purposes of comparison, the performance of the capitalisation-weighted market portfolio and of an equal-weighted portfolio of all 51 theoretical characteristics is also shown. Performance of big data parametric portfolio without transaction costs Market portfolio Equally weighted characteristic portfolios Significant characteristic portfolios Performance of big data parametric portfolio with transaction costs Market portfolio Equally weighted characteristic portfolios Significant characteristic portfolios 2 Source: The Sharpe ratio shows the excess portfolio return per unit of risk, with risk measured as the standard deviation of returns. 3 Source: Firm Characteristics and the Cross Section of Stock Returns: A Portfolio Perspective, De Miguel, Martin-Utrera, Nogales, Uppal, July The returns of the significant characteristic portfolio are hypothetical and based on a back-test of the portfolio construction methodology. The y-axes show cumulative historical returns from May 1989 to December 2014, rebased to 0 at the start date. Past performance is no guide to future returns.

8 ETF Research Academy 50 ways to beat the benchmark 8 Conclusions The rapidly growing interest in recent years in factor-based approaches to equity portfolio construction has led to a proliferation in academic research of potential equity market characteristics. In 2011, John Cochrane, president of the American Finance Association, spoke of a zoo of new factors and questioned how many were really needed to account for the behaviour of stock prices. DeMiguel, Martín-Utrera, Nogales and Uppal seek to answer Cochrane s question and to do so in the presence of real-world constraints, such as transaction costs and differing levels of investor risk aversion. They conclude that, from a starting list of 51 theoretical characteristics, the number of characteristics that are important for constructing optimal portfolios in the absence of transaction costs is six. In the presence of transaction costs, the number of significant characteristics increases to fifteen, because the additional characteristics help to reduce turnover and thereby constrain trading costs. Further, the researchers argue that a big data approach to portfolio construction helps identify a broader set of potential equity market characteristics, improving both on the risk-return characteristics of the standard, capitalisation-weighted market portfolio and on those of a portfolio consisting only of the most commonly used factors from academic literature.

9 ETF Research Academy 50 ways to beat the benchmark 9 About the authors Dr. Alberto Martin-Utrera Dr. Alberto Martin-Utrera is a Lecturer in Finance at Lancaster University Management School. Alberto s research focuses on the interface between asset allocation and statistics. In particular, his research addresses how investors can benefit from statistical and optimization techniques that alleviate the impact of parameter uncertainty in portfolio decisions. Alberto s research has been published in finance journals such as Journal of Financial and Quantitative Analysis and Journal of Banking and Finance. Alberto received his Ph.D. in Business Administration and Quantitative Methods from Universidad Carlos III de Madrid. During his PhD, he was a visiting researcher in the Operations and Management Science department at London Business School. He holds a B.A. degree in Economics from Universidad Autonoma de Madrid, Spain. Victor DeMiguel Victor DeMiguel is a Professor of Management Science at London Business School, where he has been a full-time faculty member since he joined in 2001 after earning a PhD from Stanford University. Victor s research focuses on the design, analysis, and implementation of equity portfolio strategies in the presence of taxes, transaction costs, and estimation error. One of his most popular papers is Optimal Versus Naive Diversification: How Inefficient is the 1/N Portfolio Strategy, which received the Best Paper Award from the Institute for Quantitative Investment Research and was published in The Review of Financial Studies. Victor is a multi-award winner teacher who teaches MBA and Executive Education courses on Financial Modelling and Decision Making. Francisco Javier Nogales Martín Associate Professor at Universidad Carlos III de Madrid (UC3M) since Previously, a Visiting Professor at Universidad de Castilla-La Mancha ( ), and after that ( ), a Visiting Professor (tenure-track) at the Department of Statistics of UC3M. B.S. degree in Mathematics (1995) from Universidad Autónoma de Madrid and a Ph.D. in Mathematics (2000) from UC3M. Ramon Uppal Raman Uppal is Professor of Finance at Edhec Business School. He holds a bachelors degree in Economics (Honors) from St. Stephen s College, Delhi University, and M.A., M.B.A and Ph.D. degrees from The Wharton School of the University of Pennsylvania. Prior to working at Edhec Business School, he was at London Business School and the University of British Columbia. He has held visiting positions at Catholic University (Leuven), the MIT Sloan School of Management, and the London School of Economics and Political Science. His research focuses on optimal portfolio selection and asset allocation in dynamic environments, valuation of securities in capital markets, risk management, and exchange rates. This research has been published in The Journal of Finance, The Review of Financial Studies, Journal of Economic Theory, Journal of Financial and Quantitative Analysis, Journal of International Money and Finance, and Management Science. He is currently an editor of The Review of Asset Pricing Studies and an associate editor of The Critical Finance Review. In the past, he has served as an editor of the The Review of Financial Studies and the Review of Finance, and as an associate editor of Management Science. With Piet Sercu, he is the author of the textbook, International Financial Markets and the Firm, (1995, South-Western Publishing) and of the research monograph, Exchange Rate Volatility, Trade and Capital Flows under Alternative Exchange Rate Regimes, (2000, Cambridge University Press), which received the Sanwa Monograph Award from New York University. He has taught courses on Portfolio Choice and Asset Pricing, International Financial Markets, Multinational Financial Management, Risk Management, and Corporate Finance to students in undergraduate, masters, doctoral and executive programs. He is the recipient of the Dean s Advisory Board s Outstanding Teaching Award for 1988 at The Wharton School, the Teaching Excellence Award for undergraduate teaching in 2000 at the Sauder School of Business at The University of British Columbia, the General Excellence Teaching Award for 2001/2002 at London Business School, and numerous other teaching prizes at London Business School and EDHEC Business School. Young Investigator Award for Research Excellence (UC3M) in 2010 and 2013.

10 ETF Research Academy 50 ways to beat the benchmark 10 About the ETF Research Academy The ETF Research Academy was launched in 2014 as a joint initiative between Lyxor Asset Management and the Université Paris-Dauphine House of Finance. In contrast to the abundant research on asset management in general, the Academy s objective is to help develop a research framework for topics that are specific to ETFs and indexing. What s the central theme this year? This year, the ETF Research Academy turned its focus towards the role of ETFs in portfolio management. The five research papers sponsored by the Academy explored different elements of the value chain underlying ETFs: The choice of benchmark and the design of smart indices The role of ETFs in facilitating low-cost access to different asset classes; Previous topics Last year, the Academy sponsored five research papers with a common theme of ETF liquidity. The researchers examined topics such as the link between ETF liquidity and that of the underlying basket (Calamia, Deville, Riva), ETFs and corporate bond liquidity (Sultan), the role of ETFs in intraday price discovery (Wermers), how ETF trading affects local markets (Boehmer) and the role of the primary market in determining ETF liquidity, based on a new theoretical model (Malamud). The challenges involved in the distribution of ETFs via the financial adviser community.

11 ETF Research Academy 50 ways to beat the benchmark 11 Research paper Abstract 50 Ways to Beat the Benchmark? Designing Optimally Diversified Smart Beta ETFs There is longstanding interest in identifying anomalies that can cause shares with particular characteristics to generate above-market performance. The standard approach is to use statistical analysis to identify the characteristics (or factors ) that cause certain categories of shares to perform differently to others. However, the standard factor approach ignores important constraints faced by those involved in portfolio construction and management. In particular, investors are concerned not only with potential returns but also the desire to limit risk exposures and take into account transaction costs. Standard factor indices may also be based on a value judgement about the validity and investment attractiveness of particular factors. An index designer often prejudges which factor(s) are worthy of selection, perhaps because of intuitive appeal, superior past returns or particular risk-return characteristics. In this research paper, by contrast, academics Victor DeMiguel, Alberto Martín-Utrera, Francisco Nogales and Ramon Uppal apply an agnostic stance to the selection of characteristics. They use a statistical tool from the field of machine learning to select only those characteristics that have the highest explanatory power. In their paper, the researchers seek to answer three questions: 1. Which firm-level characteristics are significant for portfolio construction and why? 2. How does the answer to this question change in the presence of transaction costs? 3. How can investors exploit a large set of characteristics to improve the estimated performance of their portfolios? The researchers key conclusions are: Including transaction costs in the analysis increases, rather than decreasing the number of significant equity market characteristics. Once trading costs are taken into account, the number of significant characteristics increases from six to fifteen; Increasing the number of characteristics helps to reduce transaction costs by comparison with a portfolio of characteristics traded in isolation because the rebalancing activities associated with the characteristics are relatively uncorrelated; A big data approach to selecting characteristics can offer a risk-adjusted return 140% higher than that of the standard, market capitalisation-weighted portfolio 1. The research is of particular relevance for the designers of smart beta ETFs. Smart beta ETFs track indices which are constructed differently than standard, capitalisationweighted benchmarks. Their aim is to provide a superior risk-adjusted return to that of the market portfolio. THIS DOCUMENT IS DIRECTED AT PROFESSIONAL INVESTORS ONLY This document is for the exclusive use of investors acting on their own account and categorized either as eligible counterparties or professional clients within the meaning of Markets in Financial Instruments Directive 2004/39/EC. It is not directed at retail clients. In Switzerland, it is directed exclusively at qualified investors. This material reflects the views and opinions of the individual authors at this date and in no way the official position or advices of any kind of these authors or of Lyxor International Asset Management and thus does not engage the responsibility of Lyxor International Asset Management nor of any of its officers or employees. This research is not an offer to sell or the solicitation of an offer to buy any security in any jurisdiction where such an offer or solicitation would be illegal. It does not constitute a personal recommendation or take into account the particular investment objectives, financial situations, or needs of individual clients. Clients should consider whether any advice or recommendation in this research is suitable for their particular circumstances and, if appropriate, seek professional advice, including tax advice. Our salespeople, traders, and other professionals may provide oral or written market commentary or trading strategies to our clients and principal trading desks that reflect opinions that are contrary to the opinions expressed in this research. Our asset management area, principal trading desks and investing businesses may make investment decisions that are inconsistent with the recommendations or views expressed in this research.