Essays On Trading Strategy : A Review
I recently had the pleasure of diving into the latest book by Graham Giller, an engaging and innovative work that blends advanced financial concepts with accessible writing. This monograph-style book delves into the complexities of identifying optimal trading strategies in markets characterized by significant tail risk. It navigates through the Markowitz approach and utility-based theories with a keen analytical eye.
What stands out in Giller’s work is his comprehensive and detailed exploration of portfolio selection problems, particularly focusing on exponential utility and the Laplace return distribution. While the book notably omits an alpha-model, this seems a deliberate choice, inviting readers to engage with the material actively. This aspect turns the book into not just a source of information but a platform for practical application, encouraging readers to develop their alpha-signals and construct robust, scientifically sound trading strategies.
Another aspect of the book that resonated with me was the level of mathematical content!
Giller, a physicist by training, employs a presentation style that is both rigorous yet accessible, mirroring the clarity and approachability found in physics literature. This style particularly reminded me of Lev Okun’s textbook on quarks and leptons, where Okun aimed to create a book that could be easily read on a tram. While trams aren’t common in my area, I’m confident that Giller’s book is equally suitable for a train journey. The number of equations per page strikes the perfect balance for those familiar with elementary particle physics, aligning closely with what I consider an ideal ratio.
In conclusion, Graham Giller’s book is not just an academic achievement but a testament to his ability to make complex theories understandable and engaging. It’s a commendable work that stands as a significant contribution to the field. Congratulations to Graham Giller on this exceptional accomplishment!
-Review written by Dr. Igor Halperin
Verdict? Buy The Book! A must read!
Amazon.com: Essays On Trading Strategy (World Scientific Series In Finance): 9789811273810: Giller, Graham L
Essays on Trading Strategy : Author Statement : Graham Giller, November, 2023
When working in PDT, at Morgan Stanley in the 1990’s, I tried, many >mes, to use mean variance op>miza>on (M.V.O.) when backtes>ng trading strategies. It never worked, and was always out-performed by much simpler rules. When I set up our family office in the 2000’s, I wanted to understand why this was so. I had seen many quants, essen>ally, abandon science in favour of ad hoc rules and strategies in response to such an observa>on, but I decided that I should try to understand what was wrong about the model and not simply declare that “science works everywhere, other than in finance,” as many do. Keynes is well known for saying: “when the facts change, I change my opinions” and I interpret this to mean that when a model doesn’t match what we see in Nature the correct conclusion is that the model is wrong, not that Nature is. This was the origin of the research agenda that resulted in Essays on Trading Strategy. For many years I simply sat on these results, but more recently I decided that there was an opportunity to provide more rigorous support for the part of the investment process where an alpha becomes a posi>on, which is why I wrote the book.
One of the earliest theore>cal results was around the unreliability of the Sharpe Ra>o as a sta:s:c with which to grade the performance of both backtests and live trading. Later on I realized just how much of the “hyperparameter tuning” that Quants engage in when bringing a strategy to market is essen>ally worthless. That “improvements” to a performance of up to 0.5 (or more) Sharpe Ra>o points discovered in backtest are not meaningful at all, and this is why excessive backtes>ng is not a produc>ve ac>vity. Looking back, I realized I had probably spent many man-years of my career engaged in this fu>le endeavour.
Empirically, I discovered how remarkably stable the descrip>ons of the market provided by asymmetric GARCH with non-Normally distributed innova>ons was.1
The Generalized Error distribu>on2 seem to give an empirically accurate distribu>on and my thoughts on poraolio selec>on via U>lity Theory had me thinking about the moment genera>ng func>on of the distribu>on of returns and the impact of the higher moments on the final value of the expected U>lity integral. I saw then that such fat tailed distribu>ons were not only accurate descrip>ons of the markets but might also be the factor that explained the prior empirical result that M.V.O. doesn’t work but more “binary” discrete posi>on func>ons did. At the >me I was trading strongly correlated futures strips and my focus was on poraolio investment strategies and I was able to obtain a mul>variate decision rule that, numerically, delivered everything I wanted, but that could only be expressed in analy>cal terms more generally. It was only when actually wri>ng the book a few years ago that the closed form analy>c solu>on for a holding func>on
1 This is discussed extensively in my first book, Adventures in Financial Data Science.
2 Also frequently called the Generalized Normal distribu>on or the Exponen>al Power distribu>on.
under the Laplace distribu>on dropped out in front of me and I was so astonished that I had to work it out several >mes before I believed it was true.
Moving on from U>lity Theory to a more general stochas:c programming framework took years of thought about what the key assump>ons about alphas and posi>ons needed to be to establish a framework under which the massively mul>-horizon integrals needed to compute the expected score of a set of future behaviours could be reduced to a simpler, but s>ll challenging, homogenous policy solu>on. Those assump>ons, including that future returns, alphas and, cri>cally, posi>ons, all were stochas>c processes came slowly over >me but finally dropped into place when I was dining alone in New York ader work one day.
It seems quite remarkable to me that such complex, and in >mes challenging, mathema>cs is necessary to get to the solu>on that simple trading rules, when opera>ng in the real markets, are a beeer strategy than mean-variance op>miza>on and that this is not an ad hoc, or heuris>c, solu>on but, in fact, well supported by a theory that makes sense and that the basic answer, which is when an alpha is “large” don’t trust it as much as M.V.O. would have you do, seems completely sensible to an experienced trader.
