Too fast?

Some research was published recently by Daniel Fricke and Austin Gerig on the optimal matching times. You can read the paper, "Too Fast or Too Slow. Determining the Optimal Speed of Financial Markets", posted by one of the authors.

The original paper estimated the optimal crossing rate for typical US stocks at between 200 and 900 milliseconds and suggested that microsecond level trading was unnecessary for many, and that periodic auctions may be a good alternative to continuous trading. Like the US, the HKEx operates a continuous matching market and is subject to the same pressures for higher frequency execution, so I have run the same calculations for the HKEx using the trade data for August 2014.

Given the difference in market size it is not surprising that the HKEx optimal matching time would be slower, but it is surprising how much so. The median and mean optimal crossing interval for listings in the HSI index is 18.3 and 19.9 seconds respectively, a hundred times slower than the US. For non-HSI listings the median interval is 118.6 seconds: the mean is heavily skewed by listings which have a optimal crossing interval longer than the trading day. Below are some of the fastest and slowest times for listings in the index.

FastestInterval
(seconds)
SlowestInterval
(seconds)
0700.HK4.130322.HK43.71
0388.HK4.850066.HK43.37
0941.HK5.240291.HK42.86
2318.HK5.960144.HK41.70
0267.HK7.260023.HK37.49

The optimal time was calculated for each listing every day then averaged over the month. As in the original paper, the parameters for the optimal time calculation, (2 σ ⁄ √ω) ⁄ ψ, are statistical:

• σ, the tick size, which optimistically assumes a one tick spread for all listings
• ω, the mean trade count over five minutes
• ψ, the price volatility estimated from the last traded price each five minutes.

As in the original research, all listings are assumed to benefit from liquidity provision (correcting the interval by 1 ⁄ √3), and market correlation, although in this calculation the mean correlation, ρ, is assumed to be 0.5.