Liquidity Seeking Algorithms: How Can Alpha Expectations Influence Strategy Selection Optimisation?

By Rahul Grover, SVP in Quantitative Strategy, Jefferies and Ben Springett, Managing Director, Electronic and Program Trading, Jefferies

ben-springettLiquidity seeking algorithms (LSAs), often with an “I would” feature, provide traders with the potential to reduce risk faster than schedule- or participation-based algorithms (SBAs). Faster risk reduction is achieved through the search for outsized short term liquidity which can be accessed at low incremental cost.

The time variance of the size and source of available liquidity increases the complexity of allocation decisions within LSAs. Strategy decisions are more sensitive to factors such as expected alpha (short term) in clients’ flow, average information content in execution venues’ past executions, and the stock’s dislocation from its peers. Below we discuss how the development of customised liquidity seeking strategies is influenced by these elements.

Incorporating Client Alpha
LSAs perform best when they are customised to a client’s short term alpha. Given uncertainties in alpha duration and magnitude, it may take a few iterations to get the appropriate urgency and parameterisation into strategy selection. Clients can often rely on brokers to help estimate their short term alpha.

Brokers can use statistics such as comparison of expected-versus-realized impact, and post-execution reversion to tailor the urgency level. To avoid bias in alpha estimation, analysis should exclude orders where multiple slices are received for the same stock in the same day given the obfuscating impact this would have on the data. This analysis often results in statistically insignificant outcomes for a reasonable proportion of client flow, but when significance can be established analysis is very useful in identifying client flow subsets suitable for further discussions on strategy tuning.

Chart 1 shows how the combination of the magnitude of reversion statistics and their statistical significance can be used to identify subsets of client flow that may benefit from changing urgency levels. Allocation strategies that are tuned to this expected alpha profile can avoid either the unnecessary restraint or an indiscriminate search for liquidity.


Information Content in sourced liquidity
Large block executions provide immediate benefit to orders by significantly lowering residual risk, and therefore avoiding incremental trading cost. These executions come at the hidden cost of potentially missing a better price for some fraction of the filled shares. This hidden cost can be assessed by comparing the performance of strategies with and without access to “blocks”. In addition, proprietary analysis methods can use post-execution price movement to differentiate information content in liquidity sourced from different venues.

For block executions, it’s relevant to use a proportionately long time horizon (post execution) given the alternative to achieving a block execution would involve working the order over an extended period of time.  Venues that show a statistically significant improvement in prices after block fills can be restricted for use only by higher urgency orders. Chart 2 shows that for block executions >1% ADV, average return from execution-to-close is within one standard error for most venues. Chart 3 shows that the average cost avoided by block executions is significantly higher than the average return to the close.


Dislocation from industry/sector
Tactical allocation by LSAs to dark pools should take into account a stock’s dislocation to its industry or market. While dislocation itself is not a sufficient factor to change allocation, when it is taken in conjunction with an understanding of expected alpha and the client’s view on dislocation, it can be very useful.

As an example, if a stock is highly correlated with its industry, and the client expectation is that the correlation will persist in the short term, a LSA may underweight a short term allocation for a stock which is unfavourably dislocated. Conversely if the client does not expect the correlation to persist, or if it expects short term alpha, the LSA may continue with a search for outsized liquidity.

The ability to identify expected alpha profiles of trades should serve to enable the implementation process to be more finely tuned, thus reducing implementation costs either in terms of impact incurred or opportunity cost taken. Whilst this has traditionally been the domain only of more predictable quant driven investment strategies, we are increasingly seeing greater attention on trading catalyst and portfolio manager modelling taking place across more diversified trading desks.

Advanced execution brokers are well placed to help clients model profiles and refine order instructions / parameterisation (or to develop bespoke customisations), but only when they have garnered enough data points to draw statistically significant conclusions. The likelihood of one broker having sufficient transparency of the investment strategy origin and trade catalyst for order flow originating from a centralised dealing client is low, and thus the onus is more frequently being passed upstream to the buyside to gather and analyse the relevant datasets.

Broker contribution to the flow optimisation process can then take place as a second phase, seeking to best-fit flow being sent via more standardised algo selections designed to meet the needs of a specific subset of client flow by urgency, ADV (or another measure of difficulty to trade), time of day, etc.

Source of data for all charts is Jefferies trading universe.

We’d love to hear your feedback on this article. Please click here



recommend to friends
  • gplus
  • pinterest