Futures

Goal

A short-term forecast of the market movement for the time horizon from 10 sec up to a few minutes, or the price horizon at a distance of the few ticks from the current price, if it works, can gradually improve efficiency of the execution or market making algorithms. 

A strong “buy” signal from the price predictor can prompt an execution algo logic to be more “aggressive” on buy-side of the book while at the same time to place sell orders at some distance from the top of the book and vice versa.

The goal of the research is to train NN models for making short-term market price movement forecast and estimate efficiency of the improvement on a long out-of-sample run. We will explore classic NN architectures such as feed forward, recurrent NN(s) as well as recently suggested new architectures such as KAN networks.

The potential execution cost improvement for algo(s) embedding the price forecasting component will be estimated on a long out-of-sample market data history (5+ years)

Selected Markets:

FX: 6E (EUR/USD), 6A (AUD/USD), 6B (GBP/USD), 6N (NZD/USD), 6C (USD/CAD), 6S (USD/CHF), 6J (USD/JPY) 

Metals: GC (Gold) 

Fixed Income: 

- US Treasury: ZT (2Y), ZF (5Y), ZN (10Y) 

- Fed Funds: ZQ (30 day fed funds) 

Energy: CL (Crude Oil) 

Equity Indices: ES (S&P 500 e-mini), NQ (Nasdaq 100 e-mini) 

Agriculture: Grain: ZC (corn), ZW (wheat), Oil seeds: ZS (Soybean), ZL (Soybean Oil), ZM (Soybean meal)

The results will be published upon completion of the research.