The performance of an algorithmic trading strategy (or trading strategies) with futures contracts involve a forecast of futures price (a single futures contract) or the spread between different futures contracts

finance

Description

Title: The performance of an algorithmic trading strategy (or trading strategies) with futures contracts involve a forecast of futures price (a single futures contract) or the spread between different futures contracts (calendar spread: e.g., buy the spot month short the next month, inter-commodity spread: e.g., more bullish for china than for hong kong; buy the futures on A50 ETF on SGX and short hang seng index futures). For performance measures: for table format and measurements (simple monthly/annualized average return, volatility, sharpe ratio, success to failure ratio #right/#right+wrong, semi-variance, take away the positive or negative number in calculating the volatility) see for example Che and Fung 2006 journal of futures market / Fung and Che Hong Kong Monetary Authority working paper series for table format and measurement of performance.


1. The percentage of times your model indicates a correct / incorrect trade (whether your prediction of market direction is correct for directional trading)

2. The risk and return of the strategy: use summary statistics to describe the distribution of the payoff or returns.




Example (e.g., Hang Seng Index futures S&P 500 index futures, FTSE A50 ETF futures contracts (SGX), crude oil futures contracts, US T-bond futures contract, US T-bond futures contracts, China bond futures contracts, gold futures contract, CSI 300 contract, CAC50, DAX 30 and etc. 


Strategies: day trading a single contract or day trade a spread, overnight trade (1-day exposure) start the position today and close out tmr, spread trade: long spot month short next month or vice versa, intercommodity trade: long H index futures short hang seng  index, buy csi short S&P 500/100, russel 2000, ftse 100, nikkei 225, DAX 30


Need to know the contract specification including settlement features. 


Use at least 10-years of intradaily/daily data/weekly/monthly which provides at a minimum of 120 (monthly) observations. 


I strongly encourage you to download data from Bloomberg to show to your future employer you are Bloomberg literate, take a number of Bloomberg test at home, Bloomberg literacy tests, BAT = Bloomberg attitude test. Bloomberg literacy.



Estimation period: use first 5 year data (learning period) to construct your model, quantitatively mean square, prob of a correct prediction (up or down, or flat). Your end date of your data set should be between December 2010 and Jan 2021 (or beyond). 5 years to train your model 5 years to test the performance, 6-year training, 4 year to test the performance; if you trade once a month, then 4 year means 48 observe. Test statistical and economic performance via risk and return analyses. 


5-years passed, then you use your model to make a trade for the next month, 


Recalibrate your model parameters with each additional monthly data afterward 

Rolling window that keeps 5 years of data, increase the number of observations in your model up until 1-month before end of data period.

Stay with the same model and parameters, and use the other 5-year to test the statistical accuracy P/L distribution of the model

Estimate the distribution potential economic profit.


Instruction Files

Related Questions in finance category