Client Toxicity Scoring for RFQ Market Making

Summary

The document explores how market makers can adjust RFQ quotes for different clients. It describes tiering as a scaling factor applied after calculating bid and ask prices: clients considered less toxic may receive better prices, while clients whose trades tend to precede adverse price moves may receive worse prices. The author proposes measuring each client’s mark-to-market profit and loss a short time after trades, calculating the skewness of that distribution, and ranking clients by the result.

This is a proposed approach, not a demonstrated model. The document asks whether skewness and quantile-based classification are useful but supplies no data, results, or comparisons with alternative methods. It leaves open important design choices, including the markout horizon, sample size, controls for market conditions and instrument mix, and how toxicity scores should translate into quote adjustments. The discussion centers on RFQ market making and client-level trade outcomes.

Key ideas

• RFQ market makers can adjust quote competitiveness according to estimated client toxicity.
• The proposed score uses short-horizon mark-to-market profit and loss after each client trade.
• The author suggests ranking clients by the skewness of their trade outcome distribution.
• The document does not test the proposed score or establish that skewness predicts adverse selection.
• Markout horizon, data coverage, and controls for market conditions remain unspecified.

Tags

# Tiering value in RFQ


# Tiering value in RFQ












I was wondering what are typical strategies employed by market making firms to calculate the tiering value of each client.

So when a client create an RFQ, the market maker after calculating the BID/ASK will apply a scaling factor the tiering value depending on the client. Hence offering better prices for non-toxic clients and bad prices for toxic clients.

What are typical strategies to calculate this tiering value and hence client toxicity in RFQs?

One idea I had was to utilize historical data, mark-to-market at, let's say, 5s the PnL of all trades made for each client, and measure the skewness of that distribution. Then, we rank each client based on that skewness. If the client's skewness falls within the lower quantiles, we will consider the client as toxic; otherwise, as non-toxic. Does that type of strategy works? or there are more complex strategy and more accurate methodologies to learn that tiering value?

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.