Something we at Boosted.ai have learned from our clients is that for institutional investors, explainable AI is not a nice-to-have, it is a must. In order to pass muster for asset managers, any reason our machine learning algorithms make decisions must be transparent to their stakeholders. We like to think of our machine learning algorithms as more of a glass box than a black box. We are always improving our AI software, Boosted Insights, and are very excited to announce another new feature that greatly expands on our explainable AI.
Rankings v2 expands on our existing rankings page. We have previously discussed how our rankings are at the core of our machine learning predictions for investment management. Every stock is compared against every other one in a user’s universe by our proprietary algorithms (more on those here), using all the features the user thinks are important in stock picking. The winners and losers are then sorted into quantiles (stocks in Q1 are ranked the highest, and the ones our machine learning algos think will perform best, stocks in Q5 are ranked the lowest, and the ones our machine learning algos think will perform worst). Our rankings have always surfaced which of the features were most important in the ML decisions. Our new rankings page gives much greater granularity into the drivers behind every equity in the user’s stock universe. A user can now break down every single decision our machine learning algorithms make on a per factor basis. This allows the user to see what influences our ML’s decision to rank a stock high (or low) and make a much more informed decision about the quality of its selections. It’s explainable AI for the ML skeptic - a fundamental asset manager can, at a glance, know exactly why they agree (or disagree!) with an ML pick.
Every stock in the user’s universe now features an explain score. Every feature the user picks during the model creation process (all those they believe important in stock picking) is given an explain score for every equity in their universe. The sum of a stock’s explain score helps guide the stock’s rating (5 stars in Q1, 0 stars in Q5) and roughly matches the rank of each stock. That is, stocks with the highest explain score should also be the most highly ranked stocks. Stocks in Q1 are our 5 star stocks and top picks. These are the stocks our machine learning algorithms think will perform the best, compared to all others in the user’s universe. Stocks in Q5 are our 0 star stocks and bottoms picks. These are the stocks our machine learning algorithms think will perform the worst, compared to all others in the user’s universe. The explain scores correlate with the rankings and ratings. Stocks with the highest total explain scores will be among the highest ranks/ratings, found in Q1. Stocks with the lowest total explain scores will be among the lowest ranks/ratings, found in Q5. To illustrate, the below example of MNST shows that its explain score is +22.31 and it is ranked highest in its universe.
We’ve also added variation and dispersion figures to every equity, for even greater explainability in our AI. Presented below are the Positive & Negative drivers for MNST:
Using Return on Assets as an example - hovering over the titles brings up tooltips
Whereas our previous rankings page surfaced the top five drivers for each stock in a user’s universe, rankings v2 showcases all drivers of every stock in a user’s universe. These can be sorted, filtered and saved to PDF.
Something else we heard from our fundamental asset manager clients was that greater insight into the drivers was important to their process. We are very proud of our new security comparison screen, which showcases when our explain scores (remember, a higher explain score generally means our machine learning algorithms prefer a stock) and the actual (though normalized) data meet and also diverge. Using AMD against INTC below, we show that a user can directly compare two securities across their various positive and negative drivers. This leads to some interesting observations:
Using 60M CAPM Beta as an example:
It is up to the user to decide what to do with that information, but giving even greater insight into how our AI works is something we are impassioned about.
This is a lot to take in at once, but we wanted to share our newest advances in the realm of explainable AI and showcase for our fundamental asset manager clients the kind of deep, intensive information surfaced by Boosted Insights and our ML algorithms. Understanding both why and how AI made its decisions will help investment managers share full insights with all of their stakeholders. If you want to learn more, please reach out to us! We are happy to arrange for a demo to show you what Boosted Insights can do for your institutional investment management. Also, look out for more on the topic of explainable AI and our Rankings v2 feature in the near future.
Something we at Boosted.ai have learned from our clients is that for institutional investors, explainable AI is not a nice-to-have, it is a must. In order to pass muster for asset managers, any reason our machine learning algorithms make decisions must be transparent to their stakeholders. We like to think of our machine learning algorithms as more of a glass box than a black box. We are always improving our AI software, Boosted Insights, and are very excited to announce another new feature that greatly expands on our explainable AI.
Rankings v2 expands on our existing rankings page. We have previously discussed how our rankings are at the core of our machine learning predictions for investment management. Every stock is compared against every other one in a user’s universe by our proprietary algorithms (more on those here), using all the features the user thinks are important in stock picking. The winners and losers are then sorted into quantiles (stocks in Q1 are ranked the highest, and the ones our machine learning algos think will perform best, stocks in Q5 are ranked the lowest, and the ones our machine learning algos think will perform worst). Our rankings have always surfaced which of the features were most important in the ML decisions. Our new rankings page gives much greater granularity into the drivers behind every equity in the user’s stock universe. A user can now break down every single decision our machine learning algorithms make on a per factor basis. This allows the user to see what influences our ML’s decision to rank a stock high (or low) and make a much more informed decision about the quality of its selections. It’s explainable AI for the ML skeptic - a fundamental asset manager can, at a glance, know exactly why they agree (or disagree!) with an ML pick.
Every stock in the user’s universe now features an explain score. Every feature the user picks during the model creation process (all those they believe important in stock picking) is given an explain score for every equity in their universe. The sum of a stock’s explain score helps guide the stock’s rating (5 stars in Q1, 0 stars in Q5) and roughly matches the rank of each stock. That is, stocks with the highest explain score should also be the most highly ranked stocks. Stocks in Q1 are our 5 star stocks and top picks. These are the stocks our machine learning algorithms think will perform the best, compared to all others in the user’s universe. Stocks in Q5 are our 0 star stocks and bottoms picks. These are the stocks our machine learning algorithms think will perform the worst, compared to all others in the user’s universe. The explain scores correlate with the rankings and ratings. Stocks with the highest total explain scores will be among the highest ranks/ratings, found in Q1. Stocks with the lowest total explain scores will be among the lowest ranks/ratings, found in Q5. To illustrate, the below example of MNST shows that its explain score is +22.31 and it is ranked highest in its universe.
We’ve also added variation and dispersion figures to every equity, for even greater explainability in our AI. Presented below are the Positive & Negative drivers for MNST:
Using Return on Assets as an example - hovering over the titles brings up tooltips
Whereas our previous rankings page surfaced the top five drivers for each stock in a user’s universe, rankings v2 showcases all drivers of every stock in a user’s universe. These can be sorted, filtered and saved to PDF.
Something else we heard from our fundamental asset manager clients was that greater insight into the drivers was important to their process. We are very proud of our new security comparison screen, which showcases when our explain scores (remember, a higher explain score generally means our machine learning algorithms prefer a stock) and the actual (though normalized) data meet and also diverge. Using AMD against INTC below, we show that a user can directly compare two securities across their various positive and negative drivers. This leads to some interesting observations:
Using 60M CAPM Beta as an example:
It is up to the user to decide what to do with that information, but giving even greater insight into how our AI works is something we are impassioned about.
This is a lot to take in at once, but we wanted to share our newest advances in the realm of explainable AI and showcase for our fundamental asset manager clients the kind of deep, intensive information surfaced by Boosted Insights and our ML algorithms. Understanding both why and how AI made its decisions will help investment managers share full insights with all of their stakeholders. If you want to learn more, please reach out to us! We are happy to arrange for a demo to show you what Boosted Insights can do for your institutional investment management. Also, look out for more on the topic of explainable AI and our Rankings v2 feature in the near future.
Something we at Boosted.ai have learned from our clients is that for institutional investors, explainable AI is not a nice-to-have, it is a must. In order to pass muster for asset managers, any reason our machine learning algorithms make decisions must be transparent to their stakeholders. We like to think of our machine learning algorithms as more of a glass box than a black box. We are always improving our AI software, Boosted Insights, and are very excited to announce another new feature that greatly expands on our explainable AI.
Rankings v2 expands on our existing rankings page. We have previously discussed how our rankings are at the core of our machine learning predictions for investment management. Every stock is compared against every other one in a user’s universe by our proprietary algorithms (more on those here), using all the features the user thinks are important in stock picking. The winners and losers are then sorted into quantiles (stocks in Q1 are ranked the highest, and the ones our machine learning algos think will perform best, stocks in Q5 are ranked the lowest, and the ones our machine learning algos think will perform worst). Our rankings have always surfaced which of the features were most important in the ML decisions. Our new rankings page gives much greater granularity into the drivers behind every equity in the user’s stock universe. A user can now break down every single decision our machine learning algorithms make on a per factor basis. This allows the user to see what influences our ML’s decision to rank a stock high (or low) and make a much more informed decision about the quality of its selections. It’s explainable AI for the ML skeptic - a fundamental asset manager can, at a glance, know exactly why they agree (or disagree!) with an ML pick.
Every stock in the user’s universe now features an explain score. Every feature the user picks during the model creation process (all those they believe important in stock picking) is given an explain score for every equity in their universe. The sum of a stock’s explain score helps guide the stock’s rating (5 stars in Q1, 0 stars in Q5) and roughly matches the rank of each stock. That is, stocks with the highest explain score should also be the most highly ranked stocks. Stocks in Q1 are our 5 star stocks and top picks. These are the stocks our machine learning algorithms think will perform the best, compared to all others in the user’s universe. Stocks in Q5 are our 0 star stocks and bottoms picks. These are the stocks our machine learning algorithms think will perform the worst, compared to all others in the user’s universe. The explain scores correlate with the rankings and ratings. Stocks with the highest total explain scores will be among the highest ranks/ratings, found in Q1. Stocks with the lowest total explain scores will be among the lowest ranks/ratings, found in Q5. To illustrate, the below example of MNST shows that its explain score is +22.31 and it is ranked highest in its universe.
We’ve also added variation and dispersion figures to every equity, for even greater explainability in our AI. Presented below are the Positive & Negative drivers for MNST:
Using Return on Assets as an example - hovering over the titles brings up tooltips
Whereas our previous rankings page surfaced the top five drivers for each stock in a user’s universe, rankings v2 showcases all drivers of every stock in a user’s universe. These can be sorted, filtered and saved to PDF.
Something else we heard from our fundamental asset manager clients was that greater insight into the drivers was important to their process. We are very proud of our new security comparison screen, which showcases when our explain scores (remember, a higher explain score generally means our machine learning algorithms prefer a stock) and the actual (though normalized) data meet and also diverge. Using AMD against INTC below, we show that a user can directly compare two securities across their various positive and negative drivers. This leads to some interesting observations:
Using 60M CAPM Beta as an example:
It is up to the user to decide what to do with that information, but giving even greater insight into how our AI works is something we are impassioned about.
This is a lot to take in at once, but we wanted to share our newest advances in the realm of explainable AI and showcase for our fundamental asset manager clients the kind of deep, intensive information surfaced by Boosted Insights and our ML algorithms. Understanding both why and how AI made its decisions will help investment managers share full insights with all of their stakeholders. If you want to learn more, please reach out to us! We are happy to arrange for a demo to show you what Boosted Insights can do for your institutional investment management. Also, look out for more on the topic of explainable AI and our Rankings v2 feature in the near future.