Launch of I Know First’s AI-based Predictive System for the Brazilian Stock Market – Beating IBOVESPA

On August 1^st 2017 I Know First finished the implementation and the training period of its AI-based ranking and forecasting system for the main equities listed on the Brazilian Stock Exchange Bovespa. On this date the first Brazilian stock forecast was published for the subscribed investors in the local market. 

Since then the predictions generated returns up to twice as high as the overall local market, yielding as much as 20% in the three-month period. See Figure 1.

For each covered stock the forecasts are generated daily for 6 main time horizons, expressed in calendar days/months/years: 3 days, 7 days, 14 days, 1 month, 3 months and 1 year.

The daily updated forecast consists of two numbers: the signal which indicates the predicted direction and strength of the stock’s movement in the respective time frame, and the predictability which indicates how predictable the algorithm considers the stock’s movements to be.

I Know First covers the most liquid Bovespa stocks of the largest Brazilian publicly traded companies, with historical data of at least ca. 5 years.

Figure 1

AI Added Value to the Investors

The predictive AI system can be used by investors/traders to make smarter investment decisions by:

• identifying promising opportunities in the Brazilian stock market
• implementing custom screens as overlay to support their investment process

The AI tool is an active investment management product, suitable for institutional as well as private self-directed DIY investors, helping them outperform the market and manage their portfolios with more confidence.

Outperformance Overview

Between 1^st Aug 2017 and 1^st Nov 2017, the first three months of the forecast publication since the official launch date of the model in Brazil, I Know First’s predictive AI-engine has established a consistently benchmark-outperforming track record.

For the most predictable stocks the average trade return significantly outperformed the benchmark returns (up to factor 2) for all time frames, with performance consistently improving as higher ranked stocks were selected.

Below, ranges for the average trade return across different categories of Top 5 stocks (Long/Short and Long only) are shown for the various time horizons versus the average returns of the index (see Table 1). The table clearly shows that the stocks selected using I Know First forecasts outperform the benchmark for all time horizons.

Table 1

In order to outperform the market, it is crucial to focus on the assets which the algorithm identifies as most predictable, but then also on the ones with the strongest signals (and thus the most up/down side depending on signal’s direction).

That is why I Know First also provides the Top 10 and Top 5 Bovespa forecasts for the most predictable stocks covered.

Performance Overview in More Detail and the Importance of the Predictability Indicator

The two tables below

Table 2: unfiltered predictability (>0) and

Table 3: filtered (50% most predictable)

show in more detail the average returns per trade for the various predicted time horizons (columns) for the stocks selected using I Know First’s predictions versus the performance of the benchmark (first row, IBOV returns). It is evident that the predictability indicator is crucial to identifying consistent market outperforming opportunities and that virtually all the average returns obtained using the predictability filtered algorithmic forecasts beat the market.

Moreover, among the stocks with high predictability levels (i.e. the right table), the strongest signals on average consistently correspond to higher returns.

Also, note that not only the Long-signaled trades are outperforming the benchmark index, but also the Long/Short ones. This implies that the algorithm can successfully identify the overall market direction and good investment opportunities in both the long and short directions.

Performance Evaluation Description

All Bovespa stocks covered by I Know First and the corresponding published forecasts since the official launch of the model in Brazil 1^st Aug 2017 until 1^st Nov 2017 are taken into account.

The evaluation of the average trade returns is done in two steps:

a) first, across all those stocks covered by I Know First with Predictability levels > 0 (holds for the vast majority of the           stocks covered with very few exceptions)

b) second, after preselecting the most predictable half of them only, according to their predictabilities in each time                  frame.

In b) the preselection is done using the predictability indicator and in case of “ties” the absolute signal strength is taken into account.

In both cases a) and b) the analysis of trade returns is performed for all qualified stocks as well as for the Top 20, Top 10 and Top 5 stocks in order of absolute signal. The ranking is performed for each specific forecast time frame and date separately.

Furthermore, in each case the trade returns for 3 groups of stocks are averaged.

1. Long/Short positions, i.e. mixed positions depending on the signal direction, ranked by the absolute value of the signal
2. Long positions only, ranked by the signal value >0
3. Long positions only, ranked by the signal value >0, but only those that are among the Top ranked mixed positions. This means that if on a certain day all strongest (Top 20/10/ or 5) signals are bearish, no Long positions are qualified.

Returns are calculated for all published and qualified forecasts (see above) in the period of 01/08/2017 – 01/11/2017 as buy/sell & hold returns in the direction of the published forecast and then averaged within each category and time frame.

To compare the results to a benchmark the same evaluation was done for the Ibovespa IBOV – the main Brazilian stock index – assuming long positions for respective holding periods.

Since the forecasting time frames are given in calendar days, Table 4 details the corresponding holding periods in trading days used:

Conclusion

I Know First AI-based predictive system considers the markets holistically – it searches for patterns and relationships/interconnections in huge sets of historical daily updated, structured capital markets data and considers the financial world as a large complex system as a whole. From the patterns learned and matched to the current market conditions, the algorithm is deriving future projections for the securities. It condenses the learned patterns into two indicators representing each asset’s forecast, allowing the investor to use the ranked predictions to identify great investment opportunities and to support his/her investment process and beat the market.

The predictability indicator, available for each security and time frame, helps to track how successful the algorithm is in learning the behavior of each individual asset and thus to focus on the most promising opportunities.

After primarily focusing on the U.S. stock market, the implementation of I Know First’s deep-learning model for the Brazilian stock market and its outstanding results since the launch in August 2017 highlight the superiority of I Know First’s approach and exemplifies the adaptability and high scalability level of the system to markets across the globe.

In the year 2018, I Know First is expanding its services to local investors in China, Japan and India.