Skip to main content

Being Data Driven - Market Making

· 3 min read
Sanjeev Sarda
Sanjeev Sarda
High Performance Developer

Being Data Driven is not the same thing as using AI or ML.

What does a Data Driven Approach to market making looking like?

DataDriven

Co-Distribution of Features​

What's the co-distribution of the features that impact your business performance and the features which impact the performance of your tech stack?

Define Technology Performance​

Identify and define features that define the performance of your tech stack, for example:

  • Roundtrip latency distribution
  • Throughput distribution
  • Failover and high availability characteristics

Define Features Impacting Technology Performance​

Identify and define features that impact the performance of your tech stack, for example:

  • Messaging latency and rates
  • Event or ordering processing roundtrip latency
  • Market data processing latency
  • Market data rates and rate of drops
  • Line latency

Define Business Performance​

Identify and define features that define your business performance, for example:

  • Pnl distribution
  • Sharpe
  • Max drawdown
  • Mark to market volatility

Define Features Impacting Business Performance​

Identify and define features that impact your business performance, for example:

  • Incoming order rates
  • Distribution of order flow features like sidedness, size
  • Distribution of market features like price volatility

Other Considerations in Trading​

Do you make most of your money during tails of those features which define the performance of your tech stack? How does your tech stack look like during tail events or during times where the features that impact your pnl are at extremes? What’s the volatility of your pnl like when your tech stack is under pressure? Making changes can also improve the sharpe or drawdown characteristics.

Duck Tails

Duck Tails

Systemic vs. Idiosyncratic Events​

Consider how idiosyncratic and systemic events relate too your pnl distribution. Do you make most of your pnl from idiosyncratic or systemic events?

Common testing strategies - identify specific days or periods for replay and see how your system and business performs.

If systemic events are a driver of your pnl or business performance, then your tech needs to handle that architecturally - over allocating in a shared tenancy environment without some kind of cloud autoscaling could lead to disaster as backlogs lead to further backlogs.

How to Improve Things​

Oprah you get a loss

Improving the Average​

The easiest way to improve your average is often by removing your worst outliers - are these associated with long outages and therefore time for volatility to go against you?

Improving the Median​

How do you improve your median? Make core changes, fundamental design changes, reducing the latency of whatever takes the most time or is the most common activity - often changes to middleware, messaging paradigm, data consistency etc have the biggest impact to the median.

The biggest way to improve is often reconsidering how we conduct business itself.

Tech and Business KPI Delta​

Is there a strong relationship between an increase in a tech KPI and the distribution of the features that define the success and performance of your business?

Model the relationship between technology KPIs and different points in the distribution of your pnl. We would expect this to look different depending on the type of trading activity being conducted.

e.g. A decrease in quote processing latency vs. P90 of your loss distribution for an active hedged quoting strategy will look very different when compared to that of a single leg momentum strat.