Papers

  • Ian Domowitz

  • Kumar Giritharan

Share

How Does a Stock Trade?

Stock-Specific Peer Group Analysis and its Application to Portfolio Liquidity

All data in this paper are sourced from ITG Inc.

ABSTRACT

We introduce an activity-based approach to trading performance comparisons, in which the common denominator is an individual security. The framework faciliates answers to three questions. Is there valuable information in peer trading comparisons if one analyzes performance on a stock-by-stock basis? Aggregated comparisons across institutions and brokers disguise important differences in the effectiveness of trading process. How does a stock trade? An answer lies in the characterization of single-stock profiles. The approach exploits institutional trading activity, combining such information with market data, and repurposing old tools with the goal of reducing institutional trading cost. How does one characterize the liquidity profile of a portfolio? The answer relies on observed liquidity, as evidenced by institutional trading activity. The solution depends on building the portfolio profile up from individual securities, and yields not only the implementation costs of liquidation, for example, but also the horizon over which this may be accomplished.

Comparative studies are a staple of transaction cost analytics. Sometimes called “peer group” analysis, the goal is simple: to provide relative rankings of buy-side firms, brokers, and traders with respect to trading performance. In the process, strengths and weaknesses are uncovered and become susceptible to improvement as part of the overall investment process. 

Peer group comparisons require the standardization of a vast amount of market and trading data, as well as decisions with respect to conceptual underpinnings. For example, does one focus on the difference between value and growth investing? On the size and nature of the institutions being compared?

Our philosophy always has been, you are what you trade. In that way of thinking, peer methodology is an activity-based system. Orders that share similar characteristics, such as market conditions, region, side, order size, and so forth, are assigned to distinct categories. The degree of differentiation across categories, such as trading in emerging as opposed to developed markets, is constrained only by the extent of data available. Comparisons are limited to trading within each activity category. 

We extend that philosophy to its logical conclusion, the security itself, and introduce stockspecific peer group analytics. While activity-based metrics are most often ranking instruments, stock-specific peer group analytics addresses an additional and often-asked question, how does a stock trade?

The framework nests the ranking requirements of a peer group. We illustrate the idea through a ranking exercise across five brokers, trading large cap stocks generally, and several individual securities in particular. Grouping stocks together for comparison purposes hides large performance differences across brokers on a stock by stock basis. The actionable implication of the analysis is obvious: to the extent possible, given the overall broker relationship, institutions may hugely benefit from a routing mechanism whose mechanics are stock specific.

The value added by stock-specific transaction comparisons, relative to current use of peer group analytics in general, is in security trading profiles. This is our answer to the question, how does a stock trade? Others’ response typically relies on a combination of market data and technical analysis applied to them. In contrast, the decision support framework enabled by stock-specific peer group analytics is a marriage of market data to actual trading experience in a security. This in turn supports repurposing of old concepts, such as relative strength indexes, to the problem of minimizing trading cost in an institutional setting. The goal is actionable information on a pre-trade and trade-monitoring basis. We clarify some of the possibilities through a case study of an individual security, ranging from expected order sizes through momentum conditions and trading strategy selection.

We apply some early returns from this paradigm by examining the liquidity characteristics of portfolios, in the spirit of the 2017 Securities and Exchange Commission’s fund liquidity program. Stock-specific analysis permits the building of liquidity profiles for portfolios in general, and ETFs in particular. The interplay between stock-specific trading costs, observed as opposed to hypothetical trading horizons, and portfolio weights, produce cost and liquidation horizons for arbitrary slices of the overall portfolio on an aggregate basis.

The Case for Stock-Specific Trading Comparisons

The heart of an equities strategy is the selection of individual stocks. Despite the inherent and sometimes conflicting complications in portfolio trading, high touch facilitation, and automated trading strategies, investment decisions are implemented stock by stock. The single stock may be the common denominator when it comes time to execute an order, but data limitations and trading desk orientation guide analysis into a diagram like that of Exhibit 1.


Exhibit 1
Comparative Trading Performance by Attribute


The box related to the security is represented by the characteristics of the stock, not the stock itself. If the stock is Biogen (BIIB), for example, the relevant data points include large cap and US-listed. Not all large cap US stocks have the same liquidity characteristics, however, and BIIB may look different from both its large cap category and other securities within that category. This is illustrated by a comparison of trading cost distributions in Exhibit 2.1


Exhibit 2
Trading Cost Distributions for Large Cap Orders of $50,000 to $150,000 in Value


Relative to its large cap universe, BIIB exhibits a low-peaked distribution of outcomes with very fat tails. While median transaction costs for BIIB are almost the same as for AAPL and AMZN, for example, the probability of a poor trading outcome is much higher. Such probabilities, as well as average costs across broad categories differ, shown in Exhibit 3.

 

Read the full report >