One of the themes that ties together my research is modelling expectations and forecasts of future, uncertain events, be they related to predicting stock or bond returns, the performance of mutual funds, or understanding the dynamics of cash flows or macroeconomic variables of interest in finance.
It is true that I tend to work on very diverse topics. Recently, my research work has looked at the following: convergence trade strategies, illustrated through trades on the spread between banking shares listed as Chinese A-shares and Hong Kong H-shares; the organisation and performance of decentralised investment management practices in the pension fund industry; the performance of mutual funds domiciled in Europe; the performance of stock return prediction models; the comparative performance of various approaches for multi-period forecasting; new methods for combining forecasts to improve accuracy and; the predictability of a variety of macroeconomic variables.
I tend to work on far too many topics, which is probably unwise and something I would not necessarily recommend to students. But for me, it comes down to my background in time-series econometrics combined with my research interests in finance. For sure, I see many overlaps between these areas and try to come up with new econometric perspectives on financial data series.
I guess most of what I do revolves around predictability and forecasting whether this involves identifying skilled fund managers or macro-forecasters who show some persistence in performance, predicting macroeconomic variables, using predictability for asset allocation decisions and investment strategies, or pricing assets since this involves predicting future cash flows and estimating their riskiness.
I usually grab the opportunity to work with the industry, central banks and supranationals, provided the project has an interesting research component and perspective. For me, academic work in general and especially economic modelling is particularly fulfilling if it can be used to improve decision making. People are not interested in a forecast per se. They are interested in forecasts only insofar as they can help them make better decisions or provide better tests of their models.
In a recent project, I worked with some economists at the European Central Bank on their European Survey of Professional Forecasters, which is a quarterly survey of expectations for the rates of inflation, real GDP growth and unemployment in the eurozone. The objective was to ascertain whether it was possible to use the survey data to identify skilled forecasters to improve forecasts. We used multiple approaches based on principal components and trimmed means, performance-based weighting, least squares estimates of optimal weights as well as Bayesian shrinkage to conclude that it was very difficult to beat the simple equally-weighted average forecast; interestingly the project led to an academic publication.
With the IMF, I evaluated the predictive accuracy of their twice-annual World Economic Outlook forecasts, compared these against those of a leading private forecaster, Consensus Economics, and investigated whether using the latter forecasts could improve the former. I found that the forecast performance of both organisations was quite similar in many respects and that there was little evidence that accounting for the information embodied in the Consensus forecasts could help significantly improve the IMF forecasts. This project also led to a publication.
I also enjoy working with private institutions. The top investment banks have groups of researchers that are as good as many excellent academic finance departments and at the same time these professionals are expected to deliver bottom-line results under severe time constraints – learning from their feedback, experience, and points of views is very useful for a researcher. People with the right qualifications can identify interesting academic ideas, validate them in practice and design innovative solutions on the back of these ideas. The distance between the academic drawing board and the launch pad for new product has never been as short as it is nowadays. Think of the launch of ETFs based on the finding of a relation between stocks’ idiosyncratic volatility and mean returns.
The course introduces students to advanced topics in predictive modelling as it applies to financial time series; it also covers how forecasts should be evaluated using either statistical or economic measures of accuracy.
While I have taught forecasting in the context of a PhD programme before, it was the first time that I had the opportunity to fully integrate forecasting with finance in a course. I spent quite a big part of the course talking about equilibrium models in which we should expect to see predictable returns and talked about the joint hypothesis problem – you can only test market efficiency in conjunction with an (asset pricing) model for how expected returns should evolve.
I structured the course around the latest research on predictability, with a focus on modelling predictability in the equity premium and I tried to give participants a good overview of the Bayesian and more classical approaches so that they are aware of the tools they can use. Among other things, we discussed why we should expect to find predictability at the business cycle frequency even if markets are efficient, how to go about quantifying the degree of predictability we can expect to find; choosing good predictor variables from literally thousands of potential predictors; using forecast combination to improve on forecast precision. We also covered the latest research on how to evaluate the performance of individual prediction models in addition to comparing the predictive accuracy of two models to decide if they are equally good or if one model dominates; such tests are very important in answering if anyone can “beat the market.” Finally, we reviewed a wealth of empirical evidence on predictability of stock returns at both short and long horizons.
Owing to competition, markets are informationally near efficient and it is really difficult to predict the returns of stocks and other securities. I call this the “one-half plus epsilon game”: Under no predictability, odds will be close to even, i.e., 50-50. Most forecasting models can hope to only offer slightly better odds, i.e. one-half plus epsilon versus one-half minus epsilon, where epsilon is a small number; the current price reflects equilibrium and the stock may go up or down from there. In this context, you have a low signal to noise ratio and estimation error becomes very important, if constraining your forecasts can help improve their accuracy, it is definitely worth investigating.
The paper puts forwards a new approach to imposing economic constraints on forecasts of the equity premium. It modifies the traditional linear forecasting model in a way that better allows the model to learn from the data. The first constraint we look at is that of positive risk premia – this is quite natural as it is difficult to imagine an equilibrium setting where risk-averse investors would hold stocks if their expected compensations were negative. The second constraint imposes a range for the Sharpe ratio: there is a zero lower bound which is identical to the equity premium restriction and an upper bound that prevents the price of risk becoming too high. The Sharpe ratio of the market portfolio is extensively used in finance and academics and investors can be expected to have strong priors about its magnitude. Yet, it has not been used in this predictive context before. Empirically, we find that these constraints reduce uncertainty about model parameters, reduce the risk of selecting a poor forecasting model, and improve both statistical and economic measures of out-of-sample forecast performance. The Sharpe ratio constraint, in particular, results in considerable economic gains.
As a side note, I am a newcomer to the faculty and did not have prior knowledge of the structure and size of the programme, I must say I am very impressed by the large-scale and genuinely international nature of the programme – getting such a number of high quality participants to do a doctoral programme in a specialised field like finance is truly a unique achievement.
The group I taught stands out by its motivation. My first job after getting my PhD from Cambridge, was teaching at Birkbeck College, a University of London institution that specialises in evening classes; the experience gave me a lot of respect for people who commit to learning while working full-time. You need to be highly motivated to do something like that, and I admire people who do it, all the more so in the context of demanding doctoral programme.
Motivation is palpable in the classroom – this is not a quiet crowd, participants are very good at giving feedback and challenging the instructor to identify the tools needed to address the issues they feel strongly about, whether these are issues they face in their work, dissertation work or both – it is exactly the type of interaction that you hope for as an instructor and researcher.
Interestingly, the dialogue that was started during the course extended outside of the classroom and despite the short duration of the elective seminar, I am having meaningful exchanges with students over email about the course material and their ideas.
For those interested in empirical research, I would emphasize the importance of creativity – try and find a topical issue that can be addressed with new methods, possibly imported from other fields, and/or relies on a new data set or a new approach to data.