Tickbooster rund automated backtest simulations each week over the last 3 months with a 2 week padding to ensure relaxation of active positions at the end of the simulated period. Before each simulation the simulator generates a parameter grid for the following main trading elements:
- Expiration (after how many days should we close a position no matter what) [
- Stop loss ratio
- Stop loss only after N days passed after opening position
- Take profit ratio
- Take any profit after N days passed after opening position
- Purchase success (simulates that you can not always purchase a stock because of certain technical or price movement issues)
- Minimum prediction strength
- Max budget per symbol
For example usually the parameter grid for gridsearch looks like this:
expiration_days: [0, 10], buy_fail_ratio: [0.2], max_budget_per_symbol: [5000, 10000], min_price: [15], max_price: [1000], stop_loss_ratio: [0, 0.02, 0.05], stop_loss_after_days: [0, 1, 10], take_profit_ratio: [0,0.01, 0.03,0.05], take_profit_after_days: [0, 1, 5, 10], min_confidence: [0.55, 0.65]
So combining the above parameter set, the outcome will be around 728 experiments. On top of that number, the automated simulator runs each set 30+ times with different random seeds for allocating the candidate symbol pool. At each iteration, there is usually more than enough stocks with strong prediction to buy (this is the candidate symbol pool), so we shuffle them before starting the trading cycle. This ensures enough variance in the simulation to indicate us a standard deviation for the yield.
You can see the simulator algorithm that utilizes tickbooster prediction with a given set of experiment parameters here.
Continuing the example above, after running 728*3ö iterations will result 21840 experiments. We order and filter automated experiments to have the best yielding with the least possible standard deviation. Also there are a set of criterias for sorting and filtering experiments in accordion to what makes a good trading strategy.