Short-term Trading: Summary of Realistic Backtests based on Daily Stock Selection

While I Know First’s short-term algorithmic predictions are often utilized for their relevance in timing mid and long-term investments, we perform realistic backtests that show how these predictions can be used to develop multiple day trading models which exploit market trends and offer high returns. In this article we present how the performances of several I Know First strategies are affected by increasingly realistic simulation environments and how these can be adjusted to adapt to the constraints encountered when live-trading.

We backtest several I Know First trading strategies from January 7th, 2016 to July 15th, 2016 for a starting capital of $ 1 million in Quantopian, an open sourced algorithmic trading development platform. Using Quantopian’s minute level data and built-in slippage and commission models we are able to simulate a more realistic trading environment and compute accurate statistics related to our strategies’ performances (see here for a more detailed explanation of Quantopian’s slippage and commission models used in the backtests). Furthermore, we analyze the effects on the backtest of shifting rebalancing times from market open to market close (O-to-C) to market open to market open (O-to-O). This O-to-O set of backtests is performed as it is more efficient in terms of commissions as the same stock if selected for several days in a row need not be re-traded, and offers advantages when scaling up to large assets under management as it gives the algorithm more time to rebalance its daily portfolio. No intraday adjustments such as limit orders, stop-losses or other rules are applied in these simulations (see here for an example of how such intraday trading rules can be used to complement I Know First’s trading strategies).

The backtest results are reported in the table below. For each strategy except the last, the results of 4 backtests are shown: O-to-C without and with slippage and commissions, and O-to-O without and with slippage and commissions. For the last strategy only results for the O-to-O backtests are shown.

For all strategies the attained returns register values significantly above the benchmark with superior trading statistics. However, as expected in most cases the addition of commission and slippage dent the strategies’ performances in the observed time-period (notice that for the O-to-O set of backtests the effect of adding slippage and commissions is very much contained compared to the O-to-C simulations), offering a more realistic picture of the achievable returns.

Strategies 1,2, and 3

The IKF_Top20_acc_cons_filter, IKF_Top20_consist_streak, and IKF_Top20_pure_cons_comb strategies are based on the shortest I Know First signal and predictability time horizon and offer very strong performances in the O-to-C simulations: total returns significantly above the benchmark, Sharpe Ratios above 3, and low betas (below 0.5), volatilities, and max drawdowns (see here for a more detailed analysis of the IKF_Top20_consist_streak strategy). However, the shift to O-to-O rebalancing without the addition of any further intraday trading rules results in decreases in their performance due to their reliance on only the short term signal.

Strategy 4

The IKF_Top20_same_direct_consist_filter_all_time_frames strategy instead relies on a combination of I Know First forecasts for various short-term time horizons. As can be seen in the table, this strategy is significantly more robust to the shift in trading times, even registering increases in total returns (from 31.60% to 42.80% in the backtests including slippage and commission costs). This shows that trading strategies for longer holding periods can be implemented by using forecasts for several short-term time horizons.

QSimulation_IKF_Top20_same_direct_consist_filter_all_time_frames_07012016_15072016

Equity line in Quantopian of the IKF_Top20_same_direct_consist_filter_all_time_frames strategy.

Strategy 5

For the IKF_Top10_trend_avg_Pred_then_avg_signal trading rule we only backtest O-to-O for a 5-day holding period. This strategy is again based on a combination I Know First forecasts for multiple short-term time horizons and would be the easiest to scale up to large assets under management as it only requires the rebalancing of 1/5th of the portfolio each day. This offers advantages with respect to trade volumes as the trades performed each day are smaller when compared to the strategies where the entire portfolio is rebalanced on a daily basis.

Conclusion

This article continues a series focused on testing the performance of trading strategies based on the I Know First indicators in more realistic backtests and shows that the strategies previously presented are robust to the addition of slippage and commission costs. Moreover, by combining I Know First’s forecasts for different time horizons trading rules can be developed which largely outperform the benchmark for longer holding periods and can thus be used to minimize commission and slippage effects.

Stock Predictions

I Know First Research

[email protected]

People who read this article also liked:

  1. An Analysis of the I Know First Signal and Predictability Indicators
  2. Stock Predictions & Short-term Trading: 42% in 7 months
  3. Daily Stock Trading Strategy Returns 42.59% in H1 2016
  4. Day Trading Strategy: An In-depth Analysis of Realistic Back-Tests
  5. I Know First Stock Forecast Algorithm

premium