Are your orders being picked off? You may be getting picked off by a counterparty that has better information about short-term price movements. If so, you are getting adversely selected.
Adverse selection is the regret of trading too early and can be measured several ways. One common way to measure adverse selection is to compare the price of a fill with the price of the market at a later point in time. For example, you can quantify adverse selection by measuring the change of the mid-point after 100 milliseconds or comparing the price of the fill with the mid-point of the bid/ask after some time has elapsed. Since each trade has two sides, when one side is being adversely selected, the other side benefits.
Being picked off usually refers to passive limit orders. The idea is that if you have a passive order booked on a trading venue, then you may be adversely selected because you aren’t fast enough to cancel your limit order ahead of a market move. Often this is due to your order being at the end of the time-priority queue in an order book. This definition and related analysis work well for small orders that may be generated by an algo. Market orders tend not to be ‘picked off’ since they are not booked, but the market can still move – either favourably or adversely – and there is a way to measure the degree to which this is happening.
For larger fills, we may want to modify the analysis. The reason for this is that often large trades are done in the ‘upstairs market’ rather than via exchange order books, so there isn’t a concern about being at the end of a queue of passive orders. Instead, the concern is about the counterparty’s information advantage as well as not knowing whether your counterparty has more shares behind, i.e., if you transact your full size of 100,000 shares to buy, they may have another 400,000 shares to sell which could very well move the market below the transaction price. If that were to happen, then you would feel regret for having traded too early since by waiting to trade you may have obtained a better price.
For both cases, the methodology used in most TCA analysis involves comparing prices between two times: the time of the fill and one or more times after the fill. This makes sense because we want to understand if we traded too early, so we need to have an idea of what later price opportunities we may have had to trade.
Is there a way to improve on the analysis?
One thing that we can do is to consider the specific liquidity characteristics of the stock we are trading. Different stocks have different liquidity characteristics, so it would be nice to use these characteristics in our analysis. Does it make sense to look at the same time interval for a highly liquid mega-cap stock as for a low-liquidity small cap stock? That is what often happens in TCA.
The key is to transform from regular time (like the seconds on a watch) to volume time. In volume time, our clock ‘ticks’ based on transacted volume as opposed to measuring something like the number of times that a Cesium atom vibrates in an atomic clock. Making this transformation means that each security uses a bespoke clock that depends on how that security trades. Waiting 100 milliseconds to compare a mid-point price may mean the stock has traded several times (if it is highly liquid) or not at all (if it has low liquidity). In an approach that uses volume time, we would wait for a certain number of shares to execute rather than a set number of milliseconds – for example, how much did the mid-point move between the time of my fill and after 100% of my filled volume had transacted in the market following my fill?
The other thing we can do is incorporate the volatility of the stock. The volatility provides us with a metric that characterizes the typical price fluctuations we can expect. We know intuitively that some stock prices fluctuate more than others – these fluctuations are what volatility measures.
A stock that is highly volatile will, by definition, have larger amplitude price fluctuations than a stock with a low volatility. We would expect to see more adverse selection for stocks with higher volatility when measuring using a standard clock, but it is harder to know whether it is because our counterparty had better info or if it is due to factors unrelated to our order. We can therefore normalize the calculation of adverse selection using the volatility of the stock.
By transforming to volume time and normalizing by volatility, we can have a much more tailored approach to adverse selection allowing us to aggregate across stocks that have very different liquidity characteristics. We just need to include a couple more liquidity metrics to make some simple transformations that will make it more appropriate to compare stocks with different liquidity. This could be two different stocks, or it could be the same stock in different market conditions. This is an important consideration, especially when markets enter periods of elevated volatility which can impact both volume as well as price fluctuations.
Figure 1 We show our execution of interest as a large circle at 16:20 priced at 7.75 with a volume of 25,400 shares. The red bands represent the expected price fluctuations based on the stock’s recent volatility. The green line represents the VWAP trajectory starting after our execution. The blue line shows the evolution of the mid-point. The vertical line shows the time when 25,400 shares have executed after our execution – this is the time that it took for an amount of volume equal to our fill size to execute in the market.
In Figure 1 above, we plot volatility bands in red that show the expected range of price fluctuations based on measurements of the stock’s recent volatility characteristics. The green curve represents the VWAP starting immediately after the time of our execution. The blue line shows the variation of the mid-point. The vertical line shows the time when the volume traded following the execution is equal to the volume of the execution. In this case we are buying 25,400 shares of WEED at 16:20 UTC, so the vertical line shows the time when a further 25,400 shares of WEED executed after our fill. We can then compare the VWAP (and/or the mid-point) with the volatility bands. If the VWAP line is outside the volatility bands, we would call this adverse selection. If the VWAP line is below the volatility bands when we reach the vertical line, then the buyer is adversely selected and vice versa.
This is another example of using relative metrics rather than absolute metrics in analyzing trade performance. When we use relative metrics, we consider the prevailing liquidity environment when the trades were executed. Instead of using the same time intervals for all stocks to measure adverse selection, we use intervals that are specific to each stock and the size of our execution. We use the stock-specific intraday volatility to determine if there is adverse selection. We are using metrics that consider the specific liquidity characteristics of each security allowing for deeper insights and the ability to aggregate across executions where the executions are consummated in very different liquidity environments.