Sizing a Crypto Market Maker from Trade Flow and Recent P&L

Summary

The document describes a trend-following market maker trading XBTUSD in a highly volatile market. Its current sizing method allocates leverage using a Kelly-style weighting based on recent profit and loss, reducing allocation after poor performance. The trader is seeking a way to size bids and asks based on the availability of uninformed trades, which are described as a source of short-horizon profits.

One tested signal estimates the recent rate of uninformed trading from the volume of trades moving against the quote by a chosen number of ticks. The author reports that this measure reacts late to market conditions: it reduced profit-and-loss variation but also lowered overall return on investment. The document contains no responses, comparative results, or recommended sizing rule, so it serves mainly as a description of a sizing problem and an observed limitation of a lagging flow measure. It offers no evidence that the Kelly-style method or an alternative is optimal.

Key ideas

• The described market maker uses recent profit and loss to adjust its leverage allocation with a Kelly-style weighting.
• The strategy seeks short-horizon profits from uninformed trades in a volatile crypto market.
• A recent trade-volume measure intended to identify uninformed flow was observed to lag market conditions.
• The lagging measure reduced profit-and-loss variation but also lowered overall return on investment.
• The document poses the sizing question but gives no tested solution or general recommendation.

Tags

# Trade sizing with market making and the volume of uninformed trades


# Trade sizing with market making and the volume of uninformed trades












I have a trend following market maker in a highly volatile market (XBTUSD) which makes its money from uninformed trades over a short time horizon. Sizing my bids/asks has always caused me to question my methodology. Currently, I'm just using a fixed leverage allocation weighted by the kelly criterion over a window of recent P/L such that it adjusts downwards in periods of bad performance.

I have experimented with measuring the current 'rate' of uninformed trades (which I took to be the volume/s of trades going in the wrong direction by some number of ticks, over a recent window of trades), but found that it lags behind the market, resulting in less variance P/L, but overall less ROI.

What is the best practice here?

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.