We do trade frequently (see our historical allocations here). And, almost as frequently we get asked about the value and necessity of the small “zig-zag” trades we do. Intuitively, the short answer is yes, we do think these small trades are worth the cost and effort. Most of the small trades are driven by our short term models and reflect our views about short term market direction (between 1-day and 1-week). Out of the six models currently traded in our market-timing ensemble, four have horizons of less than 1-month or no forecasting horizon at all. If we did not think these models enhanced our performance we would not include them in the ensemble in the first place. However, as the popular scientific cliché urges us: “In God we trust. Others must provide data”, we decided to look at the data.
We designed a series of simple experiments to test the value of trading with as little rounding as we currently do (we round to the nearest 1%). These are hypothetical signals between 2003 and June 5, 2018, constructed in a way that resembles the statistical properties of our actual signals. We make realistic assumptions about trading costs: we assume we are trading E-mini futures, slipping on average half a tick relative to the last 30-minute volume weighted average price and paying $1.18 trading fees. We do not subtract management fees from these hypothetical results.
The first way to reduce trading would be to simply round all our signals to the nearest X%. The table below shows the impact of such an approach on the performance of such strategies:
There seems to be a miniscule improvement in performance when rounding to the nearest 10% but it is followed by a reduction in performance at the 20% rounding level, suggesting it is likely due to chance and not statistically significant. A quick look at the trading costs of each rounding approach shows that the trading costs actually max out at 10%.
Why don’t the costs go down as we introduce signal rounding? Since this approach has a very simple rounding rule, it leads to overtrading when the signal is near the rounding cutoffs. For example, if we were rounding to the nearest 10%, and one day our strategy suggested a 4% allocation, then 6% and then back to 4%, instead of trading 4% of our portfolio we could trade 20% (0%, to 10% and back to 0%).
Learning our lesson from simple rounding, we devised a cleverer scheme. In this scenario we will round to the nearest X% but only if the new signal is at least X% away from the old rounded signal. The tables below show the performance and costs of this approach.
This approach does help reduce trading costs. The costs associated with rounding to the nearest 10% and 20% are about 10% and 30% lower respectively. However, the savings come at a cost. We observe performance deterioration as a result of lack of response to some of the smaller signals. The performance stays about the same up to about a 5% threshold, but then we see a reduction in returns and Sharpe ratio.
In the previous approach, we focused on ignoring the small changes in the signal and only responded to more significant changes. This section takes almost an opposite approach – we will trade on all the signals below a certain threshold, but every signal above the threshold will only be executed “half way”. For example, if we are currently 10% invested, the threshold is set to 10% and the new signal tells us to go to 30%, we will only acquire a 20% exposure (the 20% is calculated as half of the signal change from current level: 20% = 10% + (30%-10%) / 2). In this case, the higher the threshold the less distorted the signal will be relative to the raw signal. We also observe this distortion empirically, by noting that the difference in performance is the most similar with the 20% threshold and becomes less and less similar and we reduce this threshold. We see a dramatic reduction in trading costs, ranging from 30% to almost 50%. However, we also give up almost two to three percentage points of returns, so we can hardly argue that the cost savings justify implementing such an algorithm.
In conclusion, we could decrease the granularity of our signal to approximately 5% increments without any significant impact on performance. We feel this is good news for anyone replicating our trading strategy using futures since the minimum number of futures required to follow our signal shrinks as we require less granularity. These experiments should also be good news for readers worried about excessive trading eating away our profits. Our conclusion from the empirical evidence shows that the smaller trades typically generate enough profit to pay for themselves.
©2018 Hull Tactical Asset Allocation, LLC (“HTAA”) is a Registered Investment Adviser.
The information set forth in HTAA’s market commentaries and writings are of a general nature and are provided solely for the use of HTAA, its clients and prospective clients. This information is not intended to be and does not constitute investment advice. The experiments described herein are for discussion and illustrative purposes only and do not reflect actual portfolio results. These materials reflect the opinion of HTAA on the date of production and are subject to change at any time without notice. Due to various factors, including changing market conditions or tax laws, the content may no longer be reflective of current opinions or positions. Past performance does not guarantee future results. All investments are subject to risks. Where data or information is presented that was prepared by third parties, such information will be cited and any such third-party sources have been deemed to be reliable. However, HTAA does not warrant or independently verify the accuracy of such information. HTAA and any third parties listed or identified herein are separate and unaffiliated, are not responsible for each other’s products, policies or services, and the views expressed are their own.