Man vs machine – the Schroders story

Share

By Jacqueline Loh, Head of Asia Trading, Schroders.
Schroders Asia first started using direct market access tools and algorithms for trade execution in 2006. Since then, the proportion of trades carried out via electronic execution has grown significantly in terms of volume of traded turnover. This would not have been possible without the increased sophistication of algos and refinement of TCA (transactions costs analysis) analytic tools to measure trading performance. In line with Schroders’ best execution policy of implementing trades with minimal total impact costs to clients, electronic trading was initially introduced to help traders minimise impact costs by gaining better overall control of their trades. It was not a conscious effort to pay less commission, although that later turned out to be one of the indirect results.
The journey over the past few years has been a very profitable and enlightening one. We made new discoveries every day, such as which broker provides the most hardy algos, which algos work best in which Asian markets, which ones were particularly suited for volatile markets. We continuously monitor the quality of our low touch and high touch execution, and the relative proportions of these to ensure that we are making optimal use of all the execution venues available to us.
The past few years have spawned a myriad of DMA and algo providers for Asia electronic trading, accompanied by a vast array of algorithms. It therefore made sense to try to differentiate between the algos as well as determine whether we were utilising algos in ways that added value.
Objective of the study: To determine if there are any significant differences in trading performance, measured against two benchmarks (arrival price and VWAP), between results achieved using fully automated algos and semi-automated algorithms.
Methodology
Using data for all of 2014, we classified all the trading strategies used by the Asia trading desk into the following two categories:
– Automated
– Semi-automated algos
Definition of automated algos: All auto schedule, auto participant type algos.
Definition of semi-automated algos: Algos with a high degree of trader input such as picking price levels and market timing.

The performance of all these algos were compared against two benchmarks – arrival price and VWAP.
Results
These are the results obtained from our two algo providers.


From the above graphs, it is clear that semi-automated algos outperform automated ones against both benchmarks – arrival price and VWAP. Divergence in performance is increasing with difficulty of trades, as represented by high ADV trades.


These results show the same trend – performance of semi-automated algos is superior with outperformance particularly marked in higher MDV type orders.
Where do we go from here
One reason for looking into this was to explore further scalability of our business i.e. if there is any deterioration in trading performance with maximum usage of algos. One Head of Execution Services had mentioned that our desk is the most scalable one he knows as we have been able to cope with large fluctuations in turnover without any change to headcount. This study shows that although efficiency can be enhanced by use of DMA tools and algorithms, the best results to the clients are achieved by a combination of trader judgement (e.g. by picking stock levels, market timing) and algorithms rather than use of fully automated algos. Differences in results are particularly marked in higher MDV orders. This has several implications for buy side trading desks. The average desk MDV (median daily volume) is a good indication of optimal algorithm usage for the desk. While desks with low average daily volume ( 20% or less as suggested by these results) may benefit from lower automation, higher average MDV desks do better with constructive trader input.
Amongst our automated algo suite, the algos we use most are the ones which enable us to accumulate or sell stock at certain levels while causing minimal market impact. Conversations with our brokers are therefore centred around refinement of their algos in such a way that allows us to trade size and yet not reveal our intent.
Automated algos too will continue to evolve to become less commoditised and more customised. We do have conversations with brokers around development of specialised, customised algos for us; however, we seem to perform better with those algos that require more manual intervention. Every desk will be different in this respect; that’s not surprising as there isn’t just one way to achieved desired outcomes.