FX Algorithms - A useful addition to your execution toolkit

In 2019, using our proprietary FX transaction cost analytics (TCA), Virtu examined the efficacy of trading using FX algos versus competitive bids.

FX Algorithms - A useful addition to your execution toolkit

In 2019, using our proprietary FX transaction cost analytics (TCA), Virtu examined the efficacy of trading using FX algos versus competitive bids. We have updated our work through October 2020 and our findings remain the same: FX algo cost distributions are much wider across all deal sizes compared to those of competitively bid trades (i.e. trades executed electronically via RFS/RFQ in a multi-dealer platform). 

As shown in Figure 5, our data revealed that the majority of algo orders were smaller than US$25M. However, as deal size increases (larger than US$10-$50M) the cost of competitive bids becomes more expensive at an increasing rate—a potential advantage for FX algos. In line with last year’s data analysis, we anticipate that the number of smaller FX algo orders may decrease as institutions become more familiar with the product; though this has not yet happened, FX algo trading is growing quickly.

In March, Greenwich Associates reported that 37% of their survey respondents used algos for 22% of their overall volume1. Six months earlier, they stated that “about one in five FX market participants are now trading via algorithm”2 showing a sharp jump in usage. JP Morgan’s FX e-commerce team reportedly told clients that 60% of orders, with a notional value of US$10M or higher, had traded algorithmically in March 20203. Within Virtu TCA, we continue to see increased adoption across our FX TCA client base (a mix of large active clients, mid-size and smaller firms).

This year, we have revisited our analysis in order to determine if our original answers on the benefits of FX algos remain true with this increasing usage. When should our clients consider using algos over competitive bids? How much more uncertainty comes when using an algo to trade? How should currency pair affect that decision?

As in 2019, we compared the cost relative to the mid-quote of trading 11 liquid-currency pairs competitively versus using FX algos. Our analysis determined:

Competitive spot prices remain priced very close to the mid-quote. 

For the most liquid pairs—EUR/USD and USD/ JPY—the median and mean performance was within -0.5 bps of the prevailing mid-quote for orders up to US$25M.

FX algos become cost-competitive when trading in size.

Once again, our analysis highlights that execution costs for competitive bids rose linearly with order size until a certain tipping point. At that point, varying from US$10M to US$50M with the currency pair, competitive order costs started increasing more rapidly — providing an opportunity for algo execution. Note that 29% of algo trades in the majors of US$25M or more outperformed the mid-quote benchmark.

In the most liquid pairs, FX algos outperformed competitive bids, but performance varied considerably more.

Just as we observed last year, algo trades had a much wider spread of outcomes compared to the competitively bid trades. 

Patience may pay off for FX algo traders. 

When adopting aggressive limits, traders could benefit from the spread capture opportunities afforded by FX algo trading. For traders using FX algos for operations purposes (i.e. to fund a trade in the equity market), FX market volatility makes these opportunities harder to capture.

FXAlgoNews
Institutional investors may have been able to reduce the cost of trading in highly volatile times by using algorithms

Methodology

Virtu’s FX TCA services over 100 clients, globally. For the purposes of this study, we compared the performance of our clients’ liquid-currency pair execution for the 12-month period of October 2019 to October 2020. Like last year, we limited the analysis to an aggregate of our largest clients—we did not want a client’s creditworthiness to affect the execution prices. In addition, we limited our analytical scope to include only trades executed electronically, either as a spot trade or the spot component of a forward trade executed via an RFS/RFQ on a multi-dealer platform or via an FX algo. Accordingly, the study does not include voice, portfolio and other trades with a manual component. We also eliminated swaps. Using the resultant data-set, we analyzed performance for the 11 currency pairs: AUD/USD, EUR/USD, GBP/USD, NZD/USD, USD/CAD, USD/CHF, USD/HKD, USD/JPY, USD/MXN, USD/SEK and USD/ ZAR. While only 5% of trades were algos, algo trades represented just over 50% of the dollar value traded in the study sample.

In the first phase of analysis, we evaluated the performance of competitive trades. In figures 2-4, we plot the distribution of performance relative to the mid-quote at the order start for three currency pairs. In each chart, we sub-divide orders into five size categories: <US$1M to >US$25M. The following box-and-whiskers plot (aka boxplot)4 charts the distribution and results from the 5th percentile to the 95th percentile performance.

From the left, the first end-point is the 5th percentile performance (worst performing), the box starts at the 25th percentile and ends at the 75th percentile while the right-most end is the 95th percentile cost. The middle is the median value.

Note: Unlike other asset classes, the distribution of FX trading costs is uniform. As a result, the weighted mean cost of all trades is approximately equal to the median cost.

How does competitively bid deal performance compare to FX algo performance?

In 2019, we had anticipated that the sample FX algo executions would be dominated by larger sizes, with little activity below US$25M. As depicted in Figure 5, for the three currency pairs, the size of the FX algo deals splits evenly into three categories: <US$10M, US$10-25M and US$25M+5. USD/MXN algo trades are slightly smaller, while the EUR/USD algo trades were larger.

Figure 6 provides an alternative view of the data, breaking down the sample by percentage executed via competitive bid versus algorithmically for the same deal sizes.

Do FX algos deliver good value?

Next, we compared the distribution of costs of competitively bid deals versus algo executed FX deals for these three order sizes. As we saw in 2019, FX algo performance data has a much wider dispersion compared to the competitive deal-bid and, for the largest deals, we begin to observe a negative skew to the results.

Other considerations

In preparing this update, we closely compared 2020 results with last year. Overall, costs increased slightly, presumably reflecting the widening of spreads during March and April. For both the whole sample and just trades in EUR/USD, USD/JPY, USD/GBP and USD/MXN, last year’s mean performance was -1.5 bps below the mid-quote. This year, the cost was -1.7 bps.

Conclusion

Institutional investors seem to be increasingly comfortable and are adding FX algos to their execution toolkit. Continuing the 2019 trend, algo trade cost distribution was much wider in comparison to competitive bids across all deal sizes. However, in deals over US$10-$50M, competitive bids reach an inflection point and become more expensive. This insight may offer a competitive advantage in the use-case for FX algos.

Conversely, our data revealed that the proportion of algo orders smaller than US$25M actually rose this year, a result that we did not anticipate. While we did not interview our clients as to why, we do have one suggestive data point. We noted earlier that the weighted mean cost of trading in EUR/USD, GBP/USD, USD/JPY and USD/MXN fell by 0.2 bps this year compared to last.

Interestingly, the cost of trading algorithmically across all sizes in those currencies remained about equal to the mid-quote prevailing at the start of the trade. In contrast, the cost for competitive bids fell from -3.2 bps to -3.9 bps. Perhaps institutional investors were able to reduce the cost of trading in highly volatile times by using algorithms. We will continue to monitor performance of FX algos and competitively bid deals for future thought leadership whitepapers.

Source:

  1. Press release: Algos in FX: Slow Start, Massive Potential published by Greenwich Associates
  2. Press release: Algos Advance on FX published by Greenwich Associates, October 22, 2019
  3. Adapt and Thrive: How FX  Algos Are Coping with Volatility. Euromoney, June 1, 2020
  4. Box and Whiskers or Boxplot: In descriptive statistics, a box plot or boxplot is a method for graphically depicting groups of numerical data through their quartiles.
  5. The smaller than expected size of FX algo orders may be due to our clients’ newness to FX algo usage. Some clients use FX algos for operational convenience, as well, even though algos carry a higher brokerage fee.