I Know First Algorithm Evaluation Report for Live ETF Universe Forecasts

Executive Summary

In this forecast evaluation report, we will examine the performance of the forecasts generated by the I Know First AI Algorithm for the ETF universe, which is sent to our customers on a daily basis. Our analysis covers the time period from 1 April 2019 to 31 May 2019. We will start with an introduction to our asset picking and benchmarking methods and then apply it to the ETF universe. We will then compare returns based on our algorithm with the benchmark performance over the same period. This evaluation is part of our continuous studies of I Know First’s live AI predictive algorithm’s performance.

About the I Know First Algorithm

The I Know First self-learning algorithm analyzes, models, and predicts the capital market, including stocks, bonds, currencies, commodities and interest rates markets. The algorithm is based on Artificial Intelligence (AI) and Machine Learning (ML), and incorporates elements of Artificial Neural Networks and Genetic Algorithms.

The system outputs the predicted trend as a number, positive or negative, along with a wave chart that predicts how the waves will overlap the trend. This helps the trader to decide which direction to trade, at what point to enter the trade, and when to exit. Since the model is 100% empirical, the results are based only on factual data, thereby avoiding any biases or emotions that may accompany human derived assumptions. The human factor is only involved in building the mathematical framework and providing the initial set of inputs and outputs to the system. The algorithm produces a forecast with a signal and a predictability indicator. The signal is the number in the middle of the box. The predictability is the number at the bottom of the box. At the top, a specific asset is identified. This format is consistent across all predictions.

Our algorithm provides two independent indicators for each asset – signal and predictability.

The signal is the predicted strength and direction of movement of the asset. This is measured from -inf to +inf.

The predictability indicates our confidence in that result. It is a Pearson correlation coefficient between past algorithmic performance and actual market movement. This is measured from -1 to 1.

You can find a detailed description of our heatmap here.

The Asset Picking Method

The method in this evaluation is as follows:

To fully utilise information provided by our forecast, we filter out the top X most predictable assets and rank them according to their predictability value. Thereafter, from them, we pick the top Y highest signals and re-adjust the rankings accordingly.

By doing so we focus on the most predictable assets on the one hand, while capturing the ones with the highest signal on the other.

For example, a top 30 predictability filter with a top 10 signal filter means that on each day we take only the 30 most predictable assets from our asset universe, and then we pick from them the top 10 assets with the highest absolute signals. On the other hand, a top 30 predictability filter with a top 30 signal filter would imply that we are solely filtering based on predictability, since we are selecting all assets in this particular set which have already been filtered by predictability.

We use absolute signals since these strategies are long and short ones. If the signal is positive, then we buy assets, i.e. open long position and, if negative, we open short position on such asset. This is to help us to identify the assets with the maximum magnitude of change, which is indiscriminate as to whether one adopts a short or long position.

The Performance Evaluation Method

We perform evaluations on the individual forecast level. This means that we calculate the return of each forecast we have issued for each horizon in the testing period. We then take the average of those results based on our positions on different assets and forecast horizon.

For example, to evaluate the performance of our 1-month forecasts, we calculate the return of each trade by using this formula:

This simulates a client purchasing the asset on the day we issue our prediction and selling it exactly 1 month in the future from that day.

We iterate this calculation for all trading days in the analyzed period and average the results.

Note that this evaluation does not take a set portfolio and follow it. This is a different evaluation method at the individual forecast level.

The Hit Ratio Calculation

The hit ratio helps us to identify the accuracy of our algorithm’s predictions.

Using our asset filtering method based on predictability and signal, we predict the direction of movement of different assets. Our predictions are then compared against actual movements of these assets within the same time horizon.

The hit ratio is then calculated as follows:

For instance, a 90% hit ratio for a top 30 predictability filter with a top 10 signal filter would imply that the algorithm correctly predicted the price movements of 9 out of 10 assets within this particular set of assets.

The Benchmarking Method

We utilise two benchmarks to determine the effectiveness of our methodology. An internal benchmark is used to determine the effectiveness of filtering assets based solely on predictability. The S&P 500 index is used as a benchmark only after filtering assets based on predictability and signal.

In the first case, the theory behind this benchmarking method is the “Null hypothesis“. This means buying every asset in the particular asset universe regardless of our I Know First indicators. For instance, if we were to identify the top 20 stock from a stock universe of 50 stocks, we would calculate the rate of return of 50 stocks, where an equal amount of each stock would be bought at the start of the time horizon and sold at the end of the time horizon. This helps us to determine the effectiveness of our predictability-based asset filtering process by comparing the rate of returns of the benchmark with the rate of return of our predictability-based strategy.

In the second case, particular stocks should be bought (or shorted) when they have been identified to have high signal strength and high predictability. We compare our rate of return based on purchasing (or shorting) the top X assets after applying both the predictability and signal filters with the rate of return of the S&P 500 index in the same time horizon. This helps us to determine the effectiveness of our methodology against the average investor.

The out-performance ratio of our trading results (based on our indicators) to benchmark results indicates the quality of the system and our indicators and provides a measure of competitive advantage an investor could get by using our forecasting solution.

Asset Universe Under Consideration: ETF Universe

In this report, we conduct testing for the ETF universe that I Know First covers in its algorithmic forecast. Our analysis covers the time period from April 1, 2019 – May 31, 2019. The asset universe under consideration is the ETF universe, or the full 140 ETFs that the I Know First forecast includes. Each daily forecast sent to I Know First’s clients includes predictions for the following time horizons: 3 days, 7 days, 14 days, 1 month, 3 month and 1 year.

Performance Evaluation: Evaluating the Predictability Indicator

We conduct our research on returns for the ETF universe for the period from 1 April 2019 to 31 May 2019. Following the methodology as described in the previous sections, we start our analysis by computing the performance of the algorithm’s long and short signals for time horizons ranging from 3 days to 1 month, without first considering the signal indicator. We apply filtering by the predictability indicator to investigate its sole marginal contribution in terms of return, and observe how returns change as these different filters are applied. Afterwards, we calculate the returns for the same time horizons for the benchmark using the stocks universe and compare it against the performance of the filtered sets of assets.

In our evaluation of the predictability indicator, the benchmark that is used is our internal benchmark, which is the returns we would have if we did not filter stocks by predictability, and merely purchased an equal amount of each stock in the Aggressive stocks universe. This is to determine the usefulness of filtering our results by our predictability indicator

Our findings are summarized in the table below:

From the above table, we can observe that generally, the marginal predictability effect provided returns that are more than two times of the benchmark returns. We further note that the best performance is recorded for Top 20 predictability stocks at the 1-month horizon, which provided returns of .49%.

Based on the above, we continue our study to identify whether the results could be improved when we filter by signal indicator.

Performance Evaluation: Evaluating the Signal Indicator

In this section, we will demonstrate how adding the signal indicator to our asset picking method improves the above performance even further. It is important to measure the performance of our strategy with respect to the benchmark, and for that we will apply the formula:

We further filter and rank the assets based on absolute signal strength to the Top 20 assets and to the Top 10 assets, which were previously already filtered by predictability.

The set of data presented is an analysis of ETFs which have previously been filtered by predictability. The data contains top 30 ETFs ranked by predictability, from which the top 20, 10, and 5 after ranking these 30 ETFs by signal.

From the above set of charts, we can clearly see that in general, if we apply signal strength filtering to the ETF universe, the Top 5 signals will both be more likely to produce greater returns than the benchmark with increasing time horizon. We see from the outperformance delta table that the 1-month return for the Top 5 signals is the best performing group, beating the benchmark with an outperformance delta of 32.85%.

For short term time horizons of 3-days and 7-days, we observe that the Top 5 signals from the was the best performing, with returns that exceeded the benchmark with an outperformance delta of 187.5% and 35.71% for the 3-days and 7-days periods respectively. This suggests that a stricter selection of ETFs has the potential to provide almost immediate returns.

Finally, when we consider the hit ratios, we see that the trends are similar to the out-performance delta trends, where we observe the peak values occurring during longer time periods. The best performing group is the Top 5 signals, which had a hit ratio of 58%. Considering that the algorithm predicted the price movements of 58% of ETFs in this sub-category and exceeded 50% for every signal and time horizon, this is very useful information for both traders and investors.

3-day, 7-day, 14-days, 1-Month, Performance: Top 30 Stocks Filtered by Predictability and Signal for ETFs Against S&P 500


In this analysis, we demonstrated the out-performance of our forecasts for ETFs picked by I Know First’s AI Algorithm during the period from 1 April 2019 to 31 May 2019 from the ETF universe. Based on the presented observations, we record significant out-performance returns of the Top 5 ETFs when our predictability and signal indicators are used together as investment criterion. As shown in the above diagram, the Top 5 ETFs filtered by predictability and signal tend to yield significantly higher returns than any other asset subset for longer time periods. Therefore, an investor who wants to critically improve the structure of his/her investments within his/her portfolio can do so by simultaneously utilizing the I Know First predictability and signal indicators as criteria for identifying the best performing stocks.

© 2019 I Know First Ltd. All Rights Reserved