This month was dedicated to incremental improvements on the technical side for three strategies: metalabeling reversals, statistical arbitrage, and funding arbitrage.

1. Reversal Strategy

The migration from the Lean framework to our Qubx framework is now mostly complete. We are in the process of validating outputs and trying to match the old simulation results from Lean with the new backtest. Below is the current result without commissions. However, this is not yet the final output and we are still investigating the differences. Once the replication of the previous baseline is finished, we will start working on more sophisticated primary strategies and features that are based on market microstructure. In the last year we have been looking into features based on orderbook and trade imbalances, combinations of open interest and funding rates, and we would like to incorporate all of this into the metalabeling strategy.

2. Statistical Arbitrage

Below is the comparison of live statistical arbitrage performance vs simulation. There are still some discrepancies and we are looking into them. The goal is to achieve parity with simulation. There might have been some issues related to maker-taker execution and live pair selection logic but this is still work in progress.

3. Funding Rate Arbitrage

Efforts this month focused on reducing execution latencies to catch better cross-exchange spreads. We have also developed an automatic capital rebalancing system that executes transfers between exchanges based on MMR requirements.

We have been testing funding arbitrage for several months live and it is becoming clear now why this risk premium still exists. There are a lot of edge cases that need to be handled properly to get this working.

  1. For instance, some small cap instrument can start pumping and the short leg will have a much larger margin requirement which can lead to low MMR on one exchange and creates a risk of liquidation.
  2. Another problem that was encountered is automatic deleveraging (ADL). This happens when a counterparty is liquidated and the position on one exchange is forcefully closed. The second leg needs to be closed immediately to remove any delta exposure and this can sometimes lead to unfavourable exit spreads.
  3. Also pair selection requires some data analysis. Simply choosing the pairs based on current funding rates is too noisy.

We have tried to tackle each of these issues and are more confident in our system. Also because of these problems we hope that this risk premium will tend to persist.

Summary

November was a month of consolidation and refinement. With the migration of reversal into Qubx in its final stages and execution improvements across statistical arbitrage and funding arbitrage underway, our infrastructure is now better positioned to support future growth.