To download all the slides from the conference is a zip file Click here
Monday 8th September 2008
09:00 James Sefton, Winton Capitals and Imperial College, London
Currently, active equity portfolio managers combine forecasting signals using static portfolio optimization procedures (all variants of mean-variance optimization). These procedures ignore all dynamic elements of portfolio construction; the most critical being the forecast horizon. How to weight a forecast based on short term price movements, with a more longer term forecast based on fundamentals?
Given investors risk-return preferences can be represented as maximizing end a power function of end of period wealth the total, return to a portfolio with regular rebalancing over a given horizon can be rewritten as an optimal risk-sensitive control problem.
Further if the dynamic evolution of the forecasts to the equity assets can be written as linear stochastic system -- which can encompass a simple representation of trading transaction costs as in Engle, Ferstenberg (2007) and Almgren, Chriss (2000) -- then the dynamic optimal portfolio can be written in terms of the solution to a matrix Riccati equation.
This optimal dynamic portfolio can be rewritten as the optimal static mean-variance portfolio plus a weighted sum of Merton (1973) hedging portfolios. This solution procedure is applied to both the forecast horizon problem described above and to the finite-horizon dynamic asset allocation problem discussed in Campbell and Viceira (2003).
10:00 Spyros Mesomeris, Citigroup Global Markets
A quantitative framework for equity market timing
In this session, Spyros will introduce a quantitative model-driven factor selection and signal diversification process that aims to generate successful recommendations to be utilized for tactical market timing in the European equity market. Using a comprehensive factor database encompassing six investment themes, Spyros will find that factor selection and forecast combination based on past predictive performance yields superior results. Spyros will also discuss dynamic factor weighting and forecast combination techniques as well as provide insights for global asset allocation.
11:30 Dan Stefek, MSCI Barra
The practice of using different factor models for risk and alpha in portfolio optimization, though widespread, has potential pitfalls. Discrepancies between risk and alpha factors can create unintended exposures in optimized portfolios that may hamper performance. An analysis reveals the root of the problem. The optimizer emphasizes the portion of the manager’s alpha that is not captured by the risk factors. Aligning the risk and alpha models may lead to better portfolios, even if doing so worsens the overall risk forecasts. Four ways of remedying these problems are presented and compared on familiar optimization problems with promising results.
13:30 Gregory Connor, London School of Economics
Presentation to be confirmed
14:30 Mark Taylor, Barclays Global Investors
The opportunity set (OS) is an ex post measure of the maximum Sharpe ratio that could have been obtained in a market for any portfolio constructed on a given universe of assets. In this paper, we show that the average value of the OS is approximately equal to the square root of portfolio breadth (number of assets) and we therefore suggest an interpretation of the OS as the “effective breadth” of a portfolio. “Effective breadth” will be higher when the volatility of asset returns is abnormally high, so that the market offered abnormal opportunities for a long-short investor—tantamount to having more portfolio breadth. Conversely, when asset returns volatility is extremely low, even the most skillful long-short investor will find it hard to generate excess returns for any given level of risk and it is as if there were little breadth at all in the portfolio. The ‘effective breadth’ interpretation still holds if we view the returns as residual to some reference portfolio that captures the general movement of the universe of assets. We provide practical examples and also derive the statistical distribution of the opportunity set through a set of Monte Carlo experiments, under a range of assumptions concerning the statistical distribution of returns. (Joint work with Richard Grinold, BGI.)
16:00 Dan di Bartolomeo, Northfield Information Services
Market events of the past year have once again focused the attention of investors, particularly quantitatively driven investors on matters of liquidity. This first part of this presentation will review the general functional form of market impact models, and on a specific functional form that offers the advantage of incorporating rational boundary conditions lacking in many similar models. We will then describe the empirical estimation of the model based on a dataset of more than 1.5 million anonymous institutional trades in liquid developed markets. The next part of the presentation will be extend the model to all traded equities into more than seventy countries including emerging and frontier markets, for which no actual trade data was available.
To do this we measure positive serial correlation in high frequency return, as a well documented manifestation of illiquidity. Using non-parametric "runs" tests, the degree of serial correlation for stocks in all the various markets is measured and mapped into the spectrum of stocks for which actual trading cost data was available. As a check of the robustness of the results, we compare our results with an entirely separately estimated market impact model based on the method Lee and Ready (1991) using tick by tick data on a sample of six thousand global equities over two years.
Tuesday 9th September 2008
09:00 Markus Leippold, Imperial College, London
We find that price and earnings momentum are pervasive features of international equity markets even when controlling for data snooping biases. For European countries, we find that price momentum is subsumed by earnings momentum on an aggregate level. However, this rationale does not apply to each and every country. While the above explanation is confined to certain time periods in the U.S., earnings momentum nevertheless appears to be a crucial driver of the price momentum anomaly in many markets. Since we cannot establish a decent relation between momentum and macroeconomic risks, we suspect a behavioral-based explanation to be at work. In fact, we find momentum profits to be more pronounced for portfolios characterized by higher information uncertainty. Hence, the momentum anomaly may well be rationalized in a model of investors under-reacting to fundamental news. Finally, we find that momentum works better when limited to stocks with high idiosyncratic risk or higher illiquidity, suggesting that limits to arbitrage deter rational investors from exploiting the anomaly.
10:00 Robert Kosowski, Imperial College, London
This paper investigates the importance of correlation risk exposure in explaining cross-sectional differences in hedge funds' performance and risk. We find that hedge funds' correlation risk exposure explains a statistically and economically significant percentage of hedge funds' absolute returns. Moreover, funds with negative loadings on the correlation risk premium have maximum drawdowns that are half as large as those of funds that sell protection against increases in correlation.
11:30 Chris Finger, RiskMetrics
Hedging credit index tranches - Investigating versions of the standard model
The standard market model for valuing credit tranches is generally assumed to be flawed, though its empirical criticisms tend to be based on static observations: the ability (or lack thereof) of the model to match the market on a given day with a small set of parameters. In practice, conventions (or abuses of the model) have partially addressed these criticisms, and the model is still used to compute hedge ratios. Here, we test the model’s performance, under a variety of conventions, in producing useful hedge ratios. We find that the more complex implementations of the model do not appear to deliver performance to warrant their complexity, and that a simple regression approach produces comparable effective hedges.
13:30 Bernd Scherer, Morgan Stanley
Portfolio construction and risk management moved in the last 10 years from a straightforward academic exercise to a big business with multiple solutions from various schools of thought and fiercely fought and entrenched battles. This contribution provides a heavily commented tour through the current practitioner thinking on the usefulness of academic concepts and widely accepted rules of thumb.
Wednesday, 10th September 2008
09:00 Giuliano De-Rossi, UBS
We look at some recent developments in the literature about the relationship between volatility and the cross section of equity returns, including the idea that highly volatile stocks underperform those with the lowest volatility levels; using long term and short term volatility as separate risk factors in a risk model; exploiting the information about variance risk premia from derivative prices and high frequency data.
10:00 Stephen Satchell, Trinity College Cambridge & BITA Risk
There is a wide perception that many of the ills of the “Quantmare” were due to failures in mean variance optimisation. From my previous research in Portfolio Skewness and Kurtosis, I have extended this work, focusing on the post “Quantmare” situation.
11:30 Edward Fishwick, BlackRock
Mechanists and Shapers: Differing quantitative responses to recent market history
The past 12 months have provided a hyper-example of the generic problems of quantitative risk and return forecasting. Rapid time variation in volatility, correlation structures, and factor structures, variation in return regimes, and high return persistence have proved problematic for many market participants. The frequent occurrence of supposedly “multi-sigma” events across a range of asset classes, the apparent inability to measure risk at all in many situations, and the difficulties of active quant strategies, are all linked in that they have undermined confidence in conventional quantitative procedures. Two broad non-exclusive responses have begun to emerge: the first involves attempts to do better quant, incorporating some of the effects described above, increasing the complexity. The second involves changing the paradigm, blending quant with “judgement”. Neither is new, both have dangers, but their relevance and importance has never been higher.