Thoughts on Crypto Market Making

In the past have been a HFT market maker for FX and other traditional instruments, however have not investigated exchange-based market making in Crypto. As I have proprietary signals applicable for Crypto, thought it would be worthwhile to investigate the difficulty of market making on crypto exchanges.

The crypto ecosystem and microstructure is quite different from FX and equities in the following ways:

• extremely high transaction costs on many exchanges
  • for example the US spot exchanges charge out at 25bps maker / taker on the lowest volume tier
  • derivatives exchanges tend to charge a somewhat more reasonable 0 - 5 bps maker / 5bps taker, more or less
  • BitMEX is an exception, providing a 2.5bps rebate for maker trades, though charges 5bps for aggressive exits.
• very tight inside spread, where the spread is 1 to 2 orders of magnitude smaller than transaction costs
  • spreads are generally in fractions of a basis point (for example 0.3 bps on BitMEX)
  • the real spread for size is quite a bit wider, however much of the volume is provided near the BBO I believe.
  • whereas maker transaction costs may generally be 0, have to factor in taker transaction costs for situations where a position needs to be aggressively liquidated.
• trades directionally on short timescales
  • observation that trading tends to move in waves where is largely buys or sells for extended periods, much more often than an even balance of buys and sells
• high volatility

Given the above, crypto MM requires a somewhat different approach. Instead of trying to capture spread, as one might do with more traditional assets, consider trying to capture somewhat longer term mean-reversion cycles.

Analysis

Examining the most recent year of tick data, did the following analysis: assuming we have attracted a long (short) position at the BBO:

• what % of the time can can one liquidate a long (short) position for a K bps profit target
  • for K = 5, 10, 25, 50, 100 bps
• unrealized drawdown one would have to ride through to liquidate for a K bps target
  • 50% and 75% quantile (drawdown 50% and 75%)
• time to liquidation
  • how long does it take on average to liquidate a position for a K bps profit (Tcleared),
• cost of failure to hedge (rfailed)
  • return if position closed after 1hr (for 5bps, 10bps) or 2hrs for 25 - 50 bps targets

5bps target

10bps target

25bps target

50bps target

Discussion

At first glance the numbers look quite attractive with 80%+ hedge achieved for targets 10bps and below; However if one considers the cost of outliers (i.e. the remaining 18% that failed to close), a naive strategy would be negative. For example the 10bps target skirts at the boundary of profitability:

81.2% x 10 - 18.8% x 46.8 -> -0.68 bps / position

With a 2.5bps rebate on BitMEX, would lead to a 2 bps+ / position profit on average, assuming passive entry and exit. However, with some additional intelligence should be able to do better.

Conditioning on various signals I have been able to increase the profitable liquidation % to up to 97%, however at the cost of much lower volume. Hence I think the approach needs to be one of:

1. a dynamic hedge model, where forecasted hedge target is variable
   • the goal would be that the larger targets would absorb the impact of the left tail
   • I have built this sort of model before, however they are non-trivial
2. conditionally enter the market passively when conditions have high probability outcomes
   • this might not be a high-volume strategy, rather very sporadic and selective
3. access bi-directional flow (possibly off exchange) to get more natural crossing
   • reducing the holding time for a position reduces the magnitude of the right tails; that said, while the tails may be smaller, the relative ratio of tail size to positive P&L can still handicap such a strategy.