Federico Bandi is currently a Professor at Johns Hopkins Carey Business School. He has been an affiliate professor in the EDHEC PhD in Finance since the beginning of the programme in 2008. He shares his expertise in the areas of financial econometrics, continuous-time asset pricing, and empirical market microstructure.
I have been working on a number of projects. Those which have more relevance for the finance community are about the measurement of the degree of illiquidity using high frequency data and the role of frequency (or the horizon) in asset pricing. Regarding the former issue, it is well-accepted that observed prices deviate from true or fundamental values. The deviations are imputed to the functioning of markets. Many would call the functioning of markets their “microstructure” and the price deviations “microstructure noise“. Much influential work in the literature has focused on learning about features of the fundamental price (for example, the volatility of the fundamental price changes) when fundamental prices are unobserved and contaminated by noise. This work effectively purges observed prices of their noise component in order to reveal fundamental counterparts. Important economic information, however, is also contained in what we have been aiming at eliminating, i.e. “noise“. If “noise“ is induced by the very nature of market operations, its study should tell us something useful about critical features of these operations, like the extent of liquidity or asymmetries in the way in which information is processed by various market participants. In work with co-authors, I show that this is indeed the case. Regarding the latter issue, I have always felt that the reason why certain canonical relations in finance do not seem to hold in the data is that they may apply to slow-moving components of the time series of interest but may be blurred when looking at the raw series because of short-term, transient effects. Think about the highly-advertised relation between the expected excess return that the market should provide with respect to riskless securities and a forecast of market variance over the same period. If variance risk is compensated, the market expected excess return should depend on market variance forecasts. It is unclear that it does so when using raw data. However, when focusing on slow-moving components of excess market return and variance series, the relation becomes considerably stronger. One could also say that the relation becomes stronger when analysing the relation at low frequencies, namely those over which a persistent signal emerges and transient perturbations have less bite.
As part of the programme’s doctoral workshop series, you presented a working paper titled “The scale of Predictability“, which studies and formalises the link between scale-specific predictability and aggregation; could you tell us about this paper and its key insights?
As I just indicated, slow-moving components of a specific process may matter for the purpose of predicting stock returns (and other finance problems) while transient dynamics may not. These slow-moving components can be filtered out. Alternatively, well-defined aggregation can be put to work as a way to extract predictive signal and reduce transient noise. The paper formalises the extraction (i.e. the filtering) of the priced slow-moving components and the role of explicit aggregation as an alternative way to conduct that extraction.
Both the work on liquidity and the work on frequency and asset pricing have uncovered interesting predictability patterns, whereby predictability I mean the ability of suitably defined variables to predict market returns over different horizons. This should be of interest to anybody engaged in asset allocation. The most popular predictor for market returns is the dividend-to-price ratio. It is well-known that this ratio should, in principle, either predict future returns or future dividend growth or both. In the data, it only seems to predict future returns. Any variable uncorrelated with the dividend-to-price ratio which is capable of predicting long run returns (and therefore adding to the predictive ability of the dividend-to-price ratio) should, instead, also predict long run dividend growth. We have identified variables that satisfy both properties (return and dividend growth predictability). These variables have the potential to add substantially to forecasting models for future market returns and lead to more effective asset allocation.
The seminar covers the estimation of nonlinear models by virtue of parametric and nonparametric methods. We discuss the mathematics of the various approaches as well as important finance applications. While the emphasis is on models written in continuous time, I first present all methods in discrete time. The methods are then adapted to a continuous-time environment by illustrating issues that are specific to the fine-grain structure of processes assumed to evolve continuously in time. I have found that the format is helpful for all constituencies of students, i.e. those with strong applied interests (whose main focus will likely be the estimation of discrete time specifications) as well as those who appear to be more drawn to theory (for whom some familiarity with continuous-time techniques appears important).
There are two main differences. First, the course structure is that of executive programmes (with a large number of hours concentrated in just a few days) and students are full-time professionals with limited amount of time outside of the classroom. Because of these realities, I tend to be a lot more detailed in class than I would be if the course allowed for (possibly) the same number of hours spread out over several weeks. The flip side of me showing every step in class is that I probably cover fewer topics than I would under a more traditional structure. My goal, however, is to give students depth in important topics and help them operationalise them in a short amount of time. The second difference has to do with the very nature of the student population. It has been an absolute pleasure teaching students with a broad range of professional experiences and achievements. I feel that I have learned from them more than I have been able to teach them.
On the contrary, I have always felt that this was the strength of the programme. I have been thoroughly impressed by the ability of the school to recruit programme participants with the right determination and technical skills to belong in a vibrant PhD programme, as well as with a rich array of professional backgrounds. This mix is rather unique and, from the view point of faculty, quite exciting.
My first piece of advice to any PhD student looking to identify a suitable topic is to work on a timely subject that makes you passionate. Research is, for the most part, a lonely endeavour, and one that often keeps you awake around the clock. Unless you are passionate about what you are working on, you are in for a rather rough ride. My second piece of advice is to place emphasis on the process, rather than on the outcome. Write your best work with passion and then think about publishing it. Publishing should not be your driver. Third, do not choose a topic just because it is an open issue in the literature. There is a reason why certain issues should remain “open”… Be selective.
If targeting top journals means writing a paper with a pre-defined idea of what should please a specific editor, I would certainly think that targeting is a bad idea. Again, very simply and somewhat trivially, we should all just be aiming at producing our best work. Once the paper’s results are in place, one may consider making adjustments to address the editorial style of a specific outlet deemed to be suitable for the research in question. This is the only form of targeting that I would advocate. In the end, while “good” papers often (not always) get into good journals, they are generally always recognised as “good”. This is all that matters. Finally, there should be an understanding that even the best editorial processes have an element of uncertainty to them. Once more, it is important to draw satisfaction from the process of writing not just from publishing. This is particularly true when working on risky projects which may increase that element of uncertainty and result in a more complex review process.