top of page
Search
Writer's pictureJeff Bacidore

Investing in Trader Alpha: A reliable, cost-effective way to boost performance

Recently, Liquidnet Founder and CEO Seth Merrin wrote an article in the Financial Times discussing how buyside trading desks can be a source of alpha.[1] Mr. Merrin’s article is concise and so well written that I’ll simply refer you to his article instead of trying to summarize it (poorly) here. In this blog post, I would like to build on this notion of how trading desks can not only add considerably to investment performance, but that the “trader alpha” they add tends to be easier to find and more persistent.


Predicting alpha vs. predicting trading costs

Anyone who has worked for an investment management firm knows how hard it is to find alpha, let alone alpha that is robust and persistent over time. This shouldn’t be surprising given the competitive nature of financial markets. Alpha is constantly at risk of “the market” uncovering that alpha and quickly “arbitraging” it away.[2] And as if things couldn’t get any worse, often when a reliable alpha is found, the “capacity” of the signal can be too limited to establish a reasonable position without trading costs eating it away. Simply put, finding alpha is hard – very, very hard.


Contrast this with trading cost prediction. Unlike alpha, trading costs tend to be more readily forecastable. For example, some brokers, fintech firms, and consultants produce “trading cost models”, which provide estimates of the expected trading costs of a given order, conditional on market conditions and order characteristics. Such tools can be extremely useful, not only in understanding the cost of a given trade, but in helping PMs – especially quant PMs – to size their trades appropriately.


Furthermore, other key intraday trading-related statistics are not only readily available, but are themselves forecastable. For example, volume, volatility, and trading costs tend to follow distinct intraday patterns.[3] And when these variables deviate from these patterns, the “surprises” themselves can be used to forecast these variables over the near term.


What does predictability have to do with performance?

Alpha is valuable only if it can be used as a predictor of future outperformance. If PMs can find assets that will outperform (or underperform), they can invest more efficiently and boost returns. Similarly, being able to predict patterns in trading costs allows traders and algorithms to trade more efficiently, by trading more when costs are low and less when costs are high. And, as noted above, trading cost predictability enables PMs to size their trades more efficiently, which also enhances investment returns.


For example, traders and algorithms can reduce their trading costs by incorporating the intraday patterns noted above when they develop their initial trading strategy (or plan). A U.S. equity trader may reduce costs by limiting how much they trade around the open, when costs tend to be at their highest, and concentrating more of their trading around the close, when costs tend to be at their lowest. And, as the execution is underway, any “surprises” in spreads, volume, volatility, etc. can be used to revise the strategy in real-time.


What about persistence?

Another important aspect of trading cost research is that its results tend to be persistent. Intraday patterns in volume, volatility, and spreads have persisted for decades, though the exact nature of the pattern has evolved (slowly) over time. Traders who exploit these patterns can consistently reduce trading costs.


And the persistence goes well beyond intraday patterns. When trading researchers uncover some cost-reducing behavior, those cost savings tend to persist as the behavior is repeated. A trading desk that switches to passive VWAP trading, for example, and finds its costs have been reduced will generally find that costs will continue to be reduced as long as the strategy is followed, assuming of course that the market environment hasn’t changed significantly. And a strategy of identifying the “best” algorithm tends to reap benefits long after the initial research has been completed. Therefore, the return on investment in trading cost research can generate significant cumulative returns, even if the per-trade savings appear to be relatively small.


How is this “trader alpha” even possible?

The “trader alpha” described above is less susceptible to arbitrage precisely because it is not really “alpha” in the same sense as investment alpha. “Trader alpha” and investment alpha are similar in that they are both measures of “value-added”. But their underlying nature is quite different. Investment alpha is an abnormal return, an anomaly of sorts. Trading costs, however, exist because of the fundamental costs associated with trading, e.g., compensation to liquidity providers for bearing risk. Trading costs tend to be high at the start of the day not because of some market inefficiency, per se. Rather, the cost of trading is higher during those times because of the higher volatility and greater adverse selection that puts liquidity providers at greater risk. And because of these increased costs, liquidity providers must charge more to remain profitable.


One way to think about this is that cost patterns are driven largely by market structure and patterns in underlying economic variables, such as intraday volatility. The former is very slow moving, generally evolving slowly over longer periods. The latter are often driven by the structural nature of the market, e.g., market openings and closing, information releases, etc. The persistence in the drivers of cost in effect help drive persistence in cost patterns.


Putting this all together

To summarize, there are three main implications of the predictability and persistence of trading cost patterns noted above. First, trading cost research can reduce trading costs due to the forecastable nature of trading costs, thereby increasing investment returns. Second, cost savings resulting from trading research are likely to persist over relatively long periods of time. In effect, the “trader alpha” can have a similar effect on performance as investment alpha, since both increase net alpha. But unlike investment alpha, trader alpha is more readily found and persistent, as it is less susceptible to being “arbitraged away”. And, third, as assets grow and trading costs become more consequential, the benefits of trading cost research actually grow with it, since the potential for cost savings increase. As assets grow, efficient trading can make the difference between positive and negative alpha capture. Contrast this with alpha, which is always at risk of reaching capacity as assets under management grow.


What to do?

Enhancing “trader alpha” simply requires firms to invest resources in trading research, aka “Trading Cost Analysis”.[4] One approach is for the buyside firm to do research to identify patterns in trading costs or to develop models of how trading costs evolve intraday. However, this is perhaps the least cost effective approach, especially when some of this information is made available by vendors and brokers.


A more cost-effective approach is to reap the benefits indirectly by doing research to find the best broker algorithm, perhaps via an “algo wheel”. Since brokers often have done the work to create models that forecast liquidity and incorporated them into their algorithms, buyside firms need only find the best performing algorithm to capture the benefits.


A related approach is to work with their brokers and consultants to customize broker algos as part of this evaluation, since as noted in a prior blog, it is unlikely that “canned” algorithm provide the optimal strategy for all of the buyside firm’s orders.[5] (And to make this blog post tax deductible, I must note that The Bacidore Group can help with any of these approaches).


But regardless, as Seth Merrin points out crisply in his article and I do more clumsily in this post, a firm’s trading desk can add measurable and consistent value to the investment process. Ignoring this is like walking past the proverbial dollar bill laying on the sidewalk without picking it up, when all this is required is a modest effort to capture that “sidewalk alpha”.



The author is the Founder and President of The Bacidore Group, LLC. For more information on how the Bacidore Group can help improve trading performance as well as measure that performance, please feel free to contact us at info@bacidore.com or by using the form at the bottom of this page.


For an overview of our other services, please click HERE.


And please check out our other blog posts available HERE.



Copyright 2019, The Bacidore Group, LLC. All Rights Reserved.



[2] I recall one such signal that arose with great strength and consistency during the months following the Lehman event. But once it found its way into a widely distributed research report, the signal vanished almost overnight (and if I recall correctly, turned negative, perhaps due to sudden overcrowding).


[3] In the U.S., for example, intraday volume tends to be J-shaped, spreads tend to form a skewed horizontal s-shape, etc. Academics and practitioners attribute the higher trading costs at the open to increased volatility and greater adverse selection risk. The lower costs at the close are typically attributed to “benign” index flows and less information asymmetry following 6+ hours of trading.


[4] Or more accurately, as Execution Strategy Analysis (ESA), as discussed in a previous blog, available here.


[5] For more information on the benefits of customization, see our blog post available here. And for a discussion of algo wheels, see our blog post here.

Comments


bottom of page