Replicating a High Frequency Trading Fund, Like for Instance, Renaissance Technologies

Since high frequency hedge funds like Renaissance Technologies always seem to be in the news I offer here a “thoughtful experiment” to tangibly explain how these firms might, I emphasize might, go about earning those stratospheric investment returns. Best of all, for anyone lacking attention span or pressed for time, I explain it all using a minimalist blog-writing style!

It is a reprint from my blog DailyScreenz: https://wordpress.com/block-editor/post/dailyscreenz.home.blog/253 which covers stock screening and other topics.

Introduction

At the top of the Mount Olympus of investment track records stands the high frequency trading firm known as Renaissance Technologies. Just by “Googling it” you’ll learn some peculiar details about this secretive, yet widely discussed, investment company. Such as, for example, how it favors hiring PHDs (mostly mathematicians and physicists) over MBAs, or its unmatched, eye popping track record (at least three decades compounding at 60% plus per year before fees), or perhaps its insanely-high fee arrangement (5% management and 44% incentive). If you Google for a few minutes more, you may read about people with low Erdős numbers and code-breaking backgrounds and other interesting academic and professional factoids. And yet, for me, the most fascinating part of this story, has always been, what gets left out: Any rational explanation of how they’ve achieved such supernatural numbers! Since no one (from the outside) seems to know enough to explain what exactly they are doing, I embarked on a financial experiment in a brazen attempt to replicate (on paper) the returns of Renaissance and others.

Understanding High Frequency Trading

High frequency funds trade in and out of thousands of stocks per day using computerized algorithms. The strategy, which seeks a small profit per trade on thousands of trades, is reliant on order flow and price volatility. I can illustrate how this process works using two “low frequency” price examples.

Table 1: Two Stock Example

Stock 1 has same open, high, low and closing prices during the trading day — so there is no change at all. With this stock, there is no room for a trader to profit by buy and selling during the session because there is no movement in price.

Stock 2 exhibits a range of prices during the trading day. In the example, we see the stock trading above and below its opening values, so there is more volatility and some potential profit to be captured by a super-fast trader stepping in and out of the order flow throughout the day. This is the basic concept that I use to create a simple high frequency replication model and if you understand this section, the rest should be intuitive.

High Frequency Replication

For my experiment, I start by selecting a stock and its daily price history (open, high, low and closing values). I then use the following process and assumptions to create a return series:

1. Calculate the daily range for each day (e.g., the difference between high and low for each day). For each trading session, I use the prior day’s range as a predictor for today’s range.
2. Assume the high frequency trading strategy can capture a percentage of today’s predicted range by trading in and out, both long and short. For the replication tests, I assumed the strategy captures 10% of each day’s estimated range as the trading profit. For example, if a stock’s high and low are $1 apart, I estimate the high frequency strategy captures $0.10 of this move on the typical day by either going long or short. It is important to note that the objective is to capture a percent of the predicted range and not the actual range for the day. For example, if the actual range for the day turned out to be $2, the strategy would still only capture $0.10, since that was the predetermined target for the day. This builds an element of sensible risk management into the process and will help the high frequency strategy from getting whipsawed.
3. Assume an error penalty, if during the trading day, the position reverses direction and goes in the wrong direction. For example, the high frequency strategy buys a stock at the open of the day, and the stock never recovers causing a loss. Or the counter situation, where a stock is sold short at the open and proceeds to rise and never comes down to generate a profit. For these instances, I assume a penalty of $0.30 or 30 cents for the days with this price behavior. I still assume some of the range is captured as described in the preceding section.
4. Assume leverage for the portfolio. For perspective, you need to realize that high frequency traders tend to set themselves up as broker dealers, providing greater access to leverage relative the average trader. I assume in the replication strategy a 7 to 1 ratio of exposure to net asset value. Put another way, for every $100 of exposure $87.5 is financed and $12.5 represents equity capital.
5. Make sure to deduct the cost of any borrowing (leverage) when calculating the daily return. I set this equal to the average 90-Day LIBOR rate for the year.

I can sit back now, and pause briefly and savor the realization that I reduced high frequency trading to just five steps and a mere 415 words! Now back to the task at hand and the question of the day: How does the replication rate in terms of returns or and does it get close to the 60% plus before fee numbers achieved by our friends?

High Frequency Replication Results

I first tested the replication on Exxon stock for the calendar year 2015. There is nothing special about Exxon or the year 2015, other than it is roughly around the time that I started thinking about simplified approaches to replicating high frequency trading returns. The results you ask, well, I’ll let you judge by the picture, but I’d say our esteemed polymaths would be appreciate the outcome!

I also tested the replication model on Exxon stock for 2016 with excellent results (a 464.7% return) but then in 2017 something strange happened. Barely any returns at all for an entire year as you can see from the next chart! When I searched for the cause it was a familiar one, and predicted earlier, a collapse in the average daily price range!

From this failure a few more lessons became apparent. Lesson 1: If you want to be a successfully high frequency shop you need to trade a broader set of stocks, and a broad enough basket that you’ll always have some stocks trading with wide daily ranges! This conclusion led me to test another mega cap stock, Microsoft. And while results were somewhat better than Exxon, a 34.7% return for 2017, it also suffered from collapsing intra-day price volatility. We still needed yet another avenue to pursue to improve our replication of the big boys.

Lesson 2: We need to also trade smaller cap companies as these may offer wider ranges to capture when large cap volatility collapses. For the next test, I looked up a small cap index and picked Zebra Technologies at random. And now we were back in business and I could appreciate why high frequency shops own all those thousands of small, mid, and large cap stocks.

Conclusion

I started thinking about the problem of finding an accessible way to test high frequency strategies using low frequency data several years ago. I put the research on the back burner and picked it up last year, dropped it, picked it up again in 2020 and decided to summarize it here. The replication approach I’ve outlined achieved all I had hoped. It provides a way for anyone with access to the Internet to test and get a feel for the most secretive investment funds around, and the approach seems to track the out-sized high frequency fund returns in an interesting way that could be extended to create benchmarking and other indicators for these strategies. And also, and most importantly, if you made it this far, thanks for reading!