CONCLUSION
E ven though this book contains an abundance of strategies that should be
interesting and attractive to independent or even institutional traders,
it has not been a recipe of strategies, or a step-by-step guide to implement-
ing them. The strategies described in this book serve only to illustrate the
general technique or concept, but they are not guaranteed to be without
those very pitfalls that I detailed in Chapter 1. Even if I were to carefully 187
scrub them of pitfalls, good strategies can still be victims of regime changes.
Readers are invited and encouraged to perform out-of-sample testing on the
strategies in this book to see for themselves.
Instead of recipes, what I hope to convey is the deeper reasons, the basic
Copyright © 2013. John Wiley & Sons, Incorporated. All rights reserved.
principles, why certain strategies should work and why others shouldn’t.
Once we grasp the basic inefficiencies of certain markets (e.g., regression
to the mean, the presence of roll returns in futures, the need for end-of-day
rebalancing in leveraged exchange-traded funds [ETFs]), it is actually quite
easy to come up with a strategy to exploit them. This notion of understand-
ing the inefficiency first and constructing a strategy later is why I empha-
sized simple and linear strategies. Why create all kinds of arbitrary rules
when the inefficiency can be exploited by a simple model?
The other notion I wanted to convey is that the approach to algorithmic
trading can be rather scientific. In science, we form a hypothesis, express it
as a quantitative model, and then test it against new, unseen data to see if
the model is predictive. If the model failed with certain data, we try to find
out the reasons for the failures, perhaps add certain variables to the model,
and try again. This is a very similar process to how we should approach
algorithmic trading. Recall the ETF pair GLD versus GDX that stopped
Chan, Ernie, and Ernest P Chan. Algorithmic Trading : Winning Strategies and Their Rationale, John Wiley & Sons,
Incorporated, 2013. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/bibliojaveriana-ebooks/detail.action?docID=1204071.
Created from bibliojaveriana-ebooks on 2022-01-16 18:15:39.
cointegrating in 2008 (see Chapter 4). A hypothesis was formed that had
to do with the high crude oil price. When oil price was added to the input
variables, the cointegration model started to work again. This scientific pro-
cess is most helpful when a strategy underperforms the backtest, and we
wanted to know why. Instead of blindly adding more rules, more indicators,
to the model and hoping that they miraculously improve the model perfor-
mance, we should look for a fundamental reason and then quantitatively test
whether this fundamental reason is valid.
Despite the efforts to make the trading process scientific and rule based,
there are still areas where subjective judgment is important. For example,
when there is a major event looming, do you trust that your model will be-
have as your backtest predicted, or do you lower your leverage or even tem-
porarily shut down the model in anticipation? Another example is offered
by the application of the Kelly formula to a portfolio of strategies. Should
we allocate capital among these strategies based on the equity of the whole
portfolio, so that the good performance of some strategies is subsidizing
the poor performance of others in the short term? Or should we apply the
Kelly formula to each strategy on its own, so that we quickly deleverage
those strategies that perform poorly recently? Mathematics tells us that the
188 former solution is optimal, but that’s assuming the expected returns and
volatilities of the strategies are unchanging. Can one really say that such ex-
CONCLUSION
pectations are unchanged given a recent period of severe drawdown?
(On the first judgment call, my experience has been that if your model
has survived the backtest during prior stressful periods, there is no reason
Copyright © 2013. John Wiley & Sons, Incorporated. All rights reserved.
to lower its leverage in the face of coming crisis. It is much better to start off
with a more conservative leverage during good times than to have to lower
it in bad ones. As Donald Rumsfeld once said, it is the “unknown unknowns”
that will harm us, not the “known unknowns.” Unfortunately, we can’t shut
down our models before the unknown unknowns strike. On the second
judgment call, my experience has been that applying Kelly to each strategy
independently so as to allow each one to wither and die quickly when it
underperforms is more practical than applying Kelly asset allocation across
all strategies.)
As these examples show, subjective judgment is often needed because
the statistical properties of financial time series are not stationary, and sci-
ence can really only deal with stationary statistics. (I am using stationary in
a sense different from the stationarity of time series in Chapter 2. Here, it
means the probability distribution of prices remains unchanged through-
out time.) Often, when we find that our live trading experience diverges
Chan, Ernie, and Ernest P Chan. Algorithmic Trading : Winning Strategies and Their Rationale, John Wiley & Sons,
Incorporated, 2013. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/bibliojaveriana-ebooks/detail.action?docID=1204071.
Created from bibliojaveriana-ebooks on 2022-01-16 18:15:39.
from the backtest, it is not because we committed any of the pitfalls during
backtesting. It is because there has been a fundamental change in the market
structure, a regime shift, due to government regulatory or macroeconomic
changes. So the fund managers still have an active ongoing role even if the
strategy is supposedly algorithmic and automated—their role is to make ju-
dicious high-level judgment calls based on their fundamental understanding
of the markets on whether the models are still valid.
However, the fact that judgment is sometimes needed doesn’t mean
that developing quantitative rules is useless or algorithmic traders are less
“smart” than discretionary traders. As the oft-quoted Daniel Kahneman
wrote, experts are uniformly inferior to algorithms in every domain that has a
significant degree of uncertainty or unpredictability, ranging from deciding
winners of football games to predicting longevity of cancer patients. One
can hope that the financial market is no exception to this rule.
189
CONCLUSION
Copyright © 2013. John Wiley & Sons, Incorporated. All rights reserved.
Chan, Ernie, and Ernest P Chan. Algorithmic Trading : Winning Strategies and Their Rationale, John Wiley & Sons,
Incorporated, 2013. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/bibliojaveriana-ebooks/detail.action?docID=1204071.
Created from bibliojaveriana-ebooks on 2022-01-16 18:15:39.
Copyright © 2013. John Wiley & Sons, Incorporated. All rights reserved.
Chan, Ernie, and Ernest P Chan. Algorithmic Trading : Winning Strategies and Their Rationale, John Wiley & Sons,
Incorporated, 2013. ProQuest Ebook Central, http://ebookcentral.proquest.com/lib/bibliojaveriana-ebooks/detail.action?docID=1204071.
Created from bibliojaveriana-ebooks on 2022-01-16 18:15:39.
