Public Information

• Home
• Events
• SFMW
• Committees
• Membership
• Student Membership
• Student Research Prize
• President's Address
• Q Links
• About our Logo

Members Only

• Membership information

Frank Ashe

Are fund management firms a corps de ballet or a bunch of prima donnas?

One common claim from a number of fund management firms is that they have a coherent house view on the world economy and markets.  One would expect that this would translate to some positive correlation between their outperformance in different asset classes.  For instance a correct view on interest rate movement should see both outperformances in the equity and fixed income portfolios, ceteris paribus.  This talk presents the result of a study to measure whether there is evidence for this in Australia.

Les Balzer

Beware the Ides (minus ten) of April -- Which Aussie equities benchmark for you?

The ASX is restructuring its indices for implementation in April 2000. The All Ordinaries Index has been deliberately redesigned to be inappropriate as a benchmark for institutional fund managers. For example, do you want to be one of the bunnies with an anachronistic All Ords benchmark chasing your share of 100% of Comalco when only 27% is not locked up in RIO's hands?

Michael Barker

Some Thoughts on Systemic Risk

We usually think of investment markets being driven by rational investors who buy or sell based on their expectations of returns. In practice, we often see overshoots, bubbles, and crashes which are then blamed on irrationality. This paper suggests that investors are often transacting in a price-insensitive fashion because of the financial situation in which they find themselves. These are "liquidity traders" rather than "portfolio traders", and they are often driven by the unwinding of leverage in some form. It is this behaviour which creates "systemic" risk.

The paper reviews some examples of past bubbles and crashes, and attempts to draw some common threads. The issue of leverage is addressed in some detail, both at the private and institutional levels, and ways of measuring, monitoring and controlling it are reviewed. Derivatives are discussed in more detail as being a particularly powerful form of leverage. Criticisms emerge of a number of conventional wisdoms.
 
 

Alan Brace

Simulating in BGM

One can simulate in BGM under several measures, Spot, Intermediate and Terminal. We compare and contrast the merits of each.
 
 

Geoffrey Brianton

Earnings revisions in the US market

Changes in consensus earnings have being recognised as having a strong relationship with future excess returns. This talk will outline some work looking at the explanatory power of earnings revision in the US Equity Market and how this can be used in the management of an equity portfolio with a "value" tilt.
 

Laurence Irlicht

Modelling Relationships which exhibit Structural Change. Can we do better than (Time) Weighted Least Squares?

Many relationships in finance are not static. Often, an asset may be strongly affected by certain factors at one point in time, and then these relationships may change - either gradually or via sudden (jump) changes. An example of the cause of a gradual change would be a company which gradually does more and more business in Asia. Its sensitivity to the Asian economies would gradually increase. Alternatively, another company may buy a gold mine, implying a sudden jump in its price correlation to gold. In cases where we do not understand the causes of asset relationships (i.e. when we only have access to the time series), and where we wish to exploit the relationships we observe, we need to be able to deduce information about the dynamic nature of those relationships. Otherwise we increase the risk of devising strategies which rely on old relationships which no longer hold. In this talk, we investigate a number of methods for the detection and analysis of changes within the relationships between assets; and discuss their practical applicability under different circumstances.
 
 

Michael Kelly and Peter Buchen

Global Generalized Linear Inverses and Generalized Binomial Trees

Generalized binomial trees are an extension of both implied and standard binomial trees. All trees can e specified in terms of their ending nodal probability distribution and a weight function that provides a recursive relation between successive nodal probabilities. Whereas the standard binomial tree has fixed up and down probabilities resulting in a lognormal ending probability distribution such that the weight function is linear, it does not allow for a volatility smile with different stikes or options with different expiration dates. These are solved with the implied and generalized binomial trees respectively, by allowing variable up and down probabilities which can describe an arbitrary terminal probability distribution and an arbitrary weight function. Thus the generalized binomial tree consistently reflects option prices on some underlying asset with both different implied volatilities and times to expiry.

The most important input to determining the tree is the asset price distribution p(T) at a fixed terminal date T. This distribution must be inferred from existing market data alone. Various methods have been proposed to estimate p(T) including: positive least squares, maximum entropy and the local generalized linear inverse method (LGLIM). In this paper we develop a new method called the "Global Generalized Linear Inverse Method" (GGLIM) to estimate p(T). This new method is an extension of the LGLIM approach. Once p(T) has been estimated, it is used to generate the entire price, volatility and expiration consistent generalized binomial tree, backwards in time from t = T to the present t = 0.

The new GGLIM approach seeks to solve the under-determined, ill-conditioned, linear inverse problem G.m = d where d is a known, noisy data vector of prices, m is a required model vector of risk-neutral probabilities and G is a known matrix which relates prices to their corresponding probabilities. A solution estimate m is sought in the form m = X. d where X is a generalized inverse of G. There is no unique X for this problem, but we can find an X which minimizes a trade-off between multiple constraints, including global resolution, information and statistical reliability measures. Such a solution is found to satisfy a well-conditioned generalised linear matrix equation depending only on known market quantities such as option and asset prices. We indicate two different methods for solving this matrix equation, which we call the spectral and kronecker methods.

Keywords Binomial trees, Linear Inverse, Optimization, Options, Kronecker product.

JEL Classification C6, G1.

Michael Kelly, Department of Mathematical Sciences, Faculty of Informatics, UWS, Macarthur, m.kelly@uws.edu.au

Peter Buchen, School of Mathematics and Statistics, Faculty of Science, University of Sydney. peterb@maths.usyd.edu.au

Terry Marsh

Estimating Factor Models of Security Returns: How Much Difference Does it Make?

Risk models for cross-sections of security returns are central to modern portfolio management, the evaluation of a portfolio manager’s performance, and to examination of how the securities are priced in relation to their risk. In this paper, we analyze the relative performance of three broad approaches to estimating these models. In the first, time series of observable macro-variables such as interest rates, growth in industrial production, and premiums on low-grade bonds, are used as proxies for realizations of the systematic risk factors affecting asset returns. In the second approach, stocks’ exposures (loosely, stocks’ “betas”) are posited to be a function of company attributes such as industrial classification, leverage, and size. The returns to the attributes in each period are then estimated by a regression across stocks of that period’s returns on the attributes. The third approach is to use the securities’ returns themselves as instruments1 for the unobservable factors and “extract” the factors from their time series.

The basic idea behind our analysis is the following: in an idealized world in which one knows the set of variables that are the true factors generating security returns, in which one can observe and measure these variables without error, and in which one knows precisely how returns depend upon the factors, then either of the first two structural approaches is an obvious choice. In all likelihood, however, the real world falls short of this ideal. A natural question is then: if, in the structural approaches, some true factors are omitted or spurious factors are introduced, if some or all of the included factors are measured with error, or an incorrect functional form is used, how substantial do the errors have to be before the third, reduced-form, approach dominates?

To answer this question, we compare the performance of the models in terms of how well an investor does in using them when trying to put together a portfolio of securities (more generally, in executing an investment strategy). Specifically, we compare the investor’s risk-adjusted expected returns, sometimes called certainty-equivalent returns, as our main basis for comparison. While we also report some goodness of fit measures and Sharpe ratio measures for added calibration, we believe that the return comparisons are the most economically relevant --- they directly incorporate the “loss function” which is usually left unspecified in conventional goodness-of-fit tests.

We find that while, in the perfect world the structural approaches outperform the third, as they must, they do not do so by a significant margin. For example, in a world2 with no measurement or specification errors, an investor with no private information (i.e. no “alphas”) and high risk aversion who construct her optimal portfolio using structural model estimates loses about 430 basis points in risk-adjusted expected return by not knowing the true risk model. If instead she had used the reduced-form model she would have lost 520 basis points because she didn’t know “truth.” On the other hand, as measurement error and other real-world problems are introduced, the performance of the structural models drops off rapidly, and they soon become dominated by the reduced form model. For example, we find that when a low-risk tolerant investor uses a structural risk model in a world in which there are 7 true factors, but two are omitted and the remaining 5 are measured perfectly, she loses about 196% of her risk-adjusted expected return if she knew truth, while she would lose only about 40% of that return using the risk model estimated from returns. In a different metric, her ex-ante Sharpe ratio measure computed from the structural models has double the error of the ratio computed from the reduced-form model. If, instead, the investor were lucky enough to include all 7 factors, but measures them with error in her structural models, she attains about 75% of true-world expected risk-adjusted expected returns using a structural model, but if instead she uses the reduced form model, she still gains about 95% of the risk-adjusted expected return.

Paul Pfleiderer

The Role of Country and Industry Effects in Explaining Global Stock Returns

We study the relative role of industry and country . In contrast to past studies, e.g. Heston and Rouwenhorst (1994) and Roll (1992), the sensitivities of stocks' returns to these factors are allowed to differ across stocks. We find that the industry factor explains 20% - 30% of the variation in stock returns that , and about 7% of the explained variation when a global factor is also included. This relative 1%-or-less estimate reported by Heston and Rouwenhorst because, we argue, they focus on . Strikingly, however, we find that the broad nine-segment "industrial classification" in the Dow Jones Global Index (DJGI) 68-industry classification, and that the 38-industry classification in the Morgan Stanley Capital International (MSCI) . Finally, we find that measuring global stock returns in a common currency decreases the return variation attributable to industry 15%.

T Sachi Purcal

Optimal Consumption, Portfolio Selection and Life Insurance for Financial Planning

This paper examines the question of lifetime personal financial planning---how should individuals determine their optimal consumption, portfolio selection and life insurance needs?  Although Richard (1975) provides the theoretical basis for such a model, no numerical results from this model have been produced.  The paper uses the Markov chain approximation method of Kushner (1977) to determine numerical results for Richard's model.  This approximation method is general, and handles constraints to the model; solutions are developed with a borrowing constraint.  The results are interpreted in light of financial planners' traditional rules of thumb for both investment in risky assets over one's lifetime and life insurance purchases.

Journal of Economic Literature Classification Numbers: C63, D91, G11, G22, J26.

Keywords: optimal portfolio selection, financial planning, life insurance, stochastic control,  Markov chain.

T. Sachi Purcal, School of Actuarial Studies, > University of New South Wales  s.purcal@unsw.edu.au

Andy Yang

Default Filter, a credit risk model for emerging markets.

An overview of a credit risk model developed specifically for emerging markets. The model was a function of the data available in the region.
 
 
 
 

Guest speakers