I Know First To Democratize AI-Driven Stock Predictions For All

I Know First Research Team LogoThis article was written by the I Know First Research Team.

The landscape of investment is a diverse one, with all sorts of actors in play – from individual investors to pension funds and other institutional investors. What unites them, among other things, is the apparent appetite that they have for the AI technology, which has recently been shaking up the financial sector. A recent poll shows that a major share of institutional investors is warming up to the boons of the high-tech era; another poll clams that consumers are ready to embrace AI in financial services. But when it comes to investment specifically, stock predictions AI algorithms trained by investment banks and other big players often end up behind the shut doors, far from the public’s eye. There are those, however, who seek to level the playing field; these include I Know First, an Israel-based AI stock predictions company. With AI having so much to offer to investors, it is up to companies like I Know First to democratize AI for all.

Stock Predictions
(Source: Pexels.com)

But how exactly would that work? How does that make the playing field a bit more level for everybody? To find out, we will take a look at how some of the largest players in the investment game utilize AI to seek alpha, and compare this with what’s on offer to the retail customers.

JPMorgan Chase: If Money Doesn’t Grow On Trees, Grow Another Random Forest

JPMorgan Chase, one of the world’s largest investment banks, is using AI for no less than stock predictions. The random forest algorithm utilized by the company is reported to deliver predictions with consistent accuracy over 67%. Its exact precision varies for the time horizon; the accuracy is also different for bullish and bearish stock predictions.

Now, given that artificial intelligence as such is perhaps more broadly associated with neural networks, it wouldn’t be a bad idea to take a step back and discuss in more depth what random forests are. To understand that, we will have to start with a simpler model – a decision tree. For the sake of the argument, let’s say we have a dataset of four variables: a customer’s employment status, age, income and education level. Based on these, we want to predict if the person defaults on a loan or not. To make things more simple, let’s view this as a binary – will vs. won’t – outcome rather than the odds of that happening.

(Source: wikiversity.org)

What a decision tree algorithm does is it finds the variable that works as the most efficient predictor and a watershed value for this variable. Let’s say it’s employment status, and a person who is employed is seen as, in most cases, non-defaulting. Then, the algorithm finds the best predictors for each of the outcomes with their respective watersheds, and this effectively comes down to a set of questions, with responses working as the basis for prediction. A random forest, in its turn, is a large collection of decision trees that utilize randomized predictors and all “vote” on the output, with the final output picked by the majority of the votes. In absolute terms, it is easier to train than a neural network, since it does not leave too much room for developers to fine-tune the algorithm. The only thing you can tweak is the number of trees, and the rule of thumb here is the more, the better. With neural networks, in the meantime, you will spend more time waiting for the AI to train, and their deployment would cost you more, since the functions are more complex than random forests.

Going back to the stock predictions AI trained by JPMorgan Chace, it is easy to see how a predictive AI with this high a precision can be of use for investors. If your portfolio is diverse enough, you are quite likely to benefit even if the AI gets 51% of the forecast right (at least as long as it doesn’t miss a plunge too deadly). Thus, no case needs to be made for how useful the total precision of 89% (6-month forecasts) can be – with a tool like this at hand, you can make your bets safely, knowing that your decision-making is based on an empirical, data-driven approach.

The catch, however, is that, in absolute terms, random forest are not necessarily the best models specifically for dealing with time series, or, in other words, with continuous processes. Granted, they can be used for that, and it is hard to make judgments without having seen the inner workings of the algorithm, but traditionally, time series analysis is where neural networks (recurrent neural networks, more specifically) thrive. The reason for that is that they are capable of picking up the seasonal patterns and making predictions based on those, while also making use of the so-called long short-term memory cells to be aware of the previous states of the system. To put it more simply, an LSTM recurrent neural network would keep track of how things are going. A random forest, however, would in most cases try to predict the future state of the system based on nothing more than the data on the current state that you fed to it and whatever it has been able to pick up in the training dataset.  

Once again, this is not to sneer at the performance of the JPMorgan Chase stock predictions model. As we noted earlier, the precision rate is more than worthy of respect, and its efficiency as a helping hand for investors is difficult to overstate. That said, it remains to be seen whether this approach would really prove to be the optimal one in the exciting field of artificial intelligence application for stock market prediction.

BlackRock: Put Your Bets On The Data

BlackRock is one of world’s largest asset management companies, and as such, it does not need a newsflash on the rise of AI in fintech. In fact, if anything, the company is setting itself up to be one of the leaders in the sphere. In 2018, it opened BlackRock Lab for Artificial Intelligence in California – a research center that is aimed at bolstering its AI capacities. Before that, it also created its Data Science Core, an initiative aimed at streamlining the efforts in data science that were already underway.

Stock Predictions
(Source: Pexels.com)

So how exactly is the company leveraging the technology in its operations? In a whole variety of ways it seems.

The first and the most important (within the framework of this article, that is) point to make is that BlackRock aims to use AI to parse through gargantuan amounts of unstructured data for insights to use in investment decisions. Here, we need to note that there has been quite a lot of talk recently about the so-called alternative data as a boon for stock market forecast AI. Alternative data is basically any data that is not generated by the company itself, be it what people say about it on social media or the number of customers walking into brick and mortar shops where its products are sold. The idea is to scrape the Web for all kinds of this data and utilize advanced neural networks to pick up meaningful patterns in those datasets, which is, it seems, precisely what BlackRock is doing.

Here, once again, we must say that the efficiency of an AI like that would depend on a whole multitude of factors, and without knowing those, it is hard to make solid assessments. But looking at this idea from the perspective of data science 101, we can note that there could be another catch in play. When we talk about deep learning as a specific approach to AI, we generally mean two things. First, our deep learning algorithm is effective a complex neural network, with one or more hidden layers between the input and output layers. In other words, if a simple neural network transforms the input into output straight away, a deep neural network would first transform it a few times (on a more entertaining note, in many cases the nature of these transformations may be mysterious to everyone but this neural network, including its creators).

Second, deep learning algorithms do not necessarily need human help in learning. With simpler algorithms, like the random forests, we would have to label all entries in the training dataset for the AI to figure out the patter. In the prior hypothetical case, for example, the training set would have to include a 5th variable, which is whether each person on the set actually defaulted or not. A deep learning AI, in its turn, can pick up patterns and correlations all by itself.

The problem is that it can also hypothetically pick up signals where there are none.

As an example, let us take one Tiger Woods, whose performance is actually surprisingly well-correlated with the fluctuations of the S&P 500 index. The consensus here is that this is nothing but law of big numbers in play: with so many golf players out there in the world, the odds that one of them would randomly mimic the stock market are not that low. However, in a hypothetical scenario where we leave our AI out in the open to figure out the best ways to predict the S&P500, we might end up having it pick up signals like this one – random correlations that may end up being as reliable as Tarot cards or any other form of divination. This is not to say that deep learning networks are bad at, let’s say, sentiment analysis – just the opposite, you can feed them a trove of tweets and get a clear picture of where things stand with the public opinion on a certain matter. But as far as other applications go, a certain degree of caution could be advised.

That said, BlackRock is steaming ahead with its push into AI as a tool for investment, and other venues – for example, its Aladdin platform is presented as an innovative AI-driven risk assessment system. As the company explores the age of AI, others are set to follow suit, and seeing who comes out on top in this competition would be as exciting as beating the markets.

Vanguard: Hybrid AI As Your Advisor

With the AI industry shaking up the financial sector, some were concerned about the disruption it was causing, wondering if the age of humans in finance was pretty much over. The reality at hand, however, is rather showcased by cases like Vanguard’s, where the AI technology is working hand in hand with human asset managers.

(Source: flickr.com)

Meet Personal Assistant Service, a robo-advisor that really, really wants to make sure that your portfolio is optimized and tailored to your financial goals and your tolerance for risk. It also just so happens to be managing around $101 billion in funds, in case you were wondering if it has been a hit with the customers. The system takes the best of both worlds by making sure that there’s both AI and human advisors working together to guarantee the most optimal outcomes for their customers.

The general idea behind such projects is to classify investors by groups, defined by their financial goals, time limits, risk appetite and other variables. From there, you can deliver tailored advice to any given customer falling into a certain category based on what other people in the same category tend to invest into. With classification as one of the two major tasks that AIs normally have to deal with, it is easy o see why this area could hold a lucrative promise for wealth managers.   

A similar approach has been picked by Schwab, the creators of the second-largest robo-advisor in terms of assets under control. We can also admire the creativity vested in their another AI-driven initiative known as Project Bear. Here, we are again talking about the classification of customers, but this time, it is all about utilizing advanced statistics in the name of behavioral analysis.

Let’s imagine the market turns bearish all of a sudden, and panic runs supreme. What would be the first instinct for a major share of investors? To exit it, of course, cutting their losses and saving what can be saved. Investing in gold, or, maybe, just going for currency. Statistically, given that the stock market would go on a rebound sooner or later, and the alternative assets may not be as safe a haven as one would imagine them to be, this may be tantamount to shooting yourself in the foot. Doing nothing, nothing at all, or even picking up some new financial instruments while everyone else is frantically dropping them off may well be a better option.

Schwab is aware of this, and the company’s AI tracks the behavior of its clients to figure out if they would be likely to follow suit in case a large sell-off starts. Those who are classified as such would get a message telling them that holding on to their assets is likely to be a better plan. This simple concept, paired with a high-tech implementation, delivers on one of the key strengths of AI-driven investment: it helps humans tone down the emotions a bit, keeping their decision-making more straight even under pressure.

Accessible High-Tech: I Know First To Democratize AI For All

While all of the above is exciting and promising, the problem is that most of this is happening behind the shut doors. The AIs trained by the large players are quite likely to end up stay unavailable for most of the actors on the market, whether we speak of other institutional investors or retail clients seeking to make their own decisions based on AI forecasts rather than put their trust in a hybrid robo-advisor. Institutional investors, in the meantime, can often find themselves facing a lack of resources, or manpower, or experience necessary to make the most out of AI-driven investment.

This is exactly where high-tech startups seeking to democratize AI for all enter the stage – and in this sphere, one of the leaders is an Israel-based company with the ambitious name I Know First.

The company has trained a deep learning stock predictions algorithm on a dataset covering 15 years of trading. Its proprietary AI looks at the market from a holistic perspective, looking for trends and patterns in the fresh market data. It models and predicts the price dynamics for more than 13,500 assets, including ETFs, stocks, currency pairs and interest rates. Drawing upon genetic programming, in goes through a learning cycle on every iteration; in other words, it keeps tracks of its own successes and failures and re-configures its models to reflect the current state of the market. This feature allows it to adapt to new situations, making sure it will not be completely dumbfounded by a pattern it had never encountered before.   

Stock Predictions
(Source: Iknowfirst.com)

The stock predictions algorithm’s output is presented as a heatmap with two indicators – signal and predictability. Signal stands for the overall performance of the asset against others on the forecast, while predictability demonstrates how well the algorithm has been able to predict the stock before. The latter is defined as the Pearson correlation coefficient between the past forecasts and the actual stock movements. With the system picking out the most predictable assets by default, customers are guaranteed to make the most out of all stock predictions get. The forecasts are delivered for a wide variety of time horizons, ranging from 3 to 365 days, to include both the best stocks to short and the optimal options for long-term investment, which gives the customer all the power and intelligence in tailoring the exact course of action to their needs.

The stock predictions AI demonstrates an impressive accuracy rate: a recent trial has seen it demonstrating an accuracy of up to 79% in predicting the dynamics of the S&P500 index. The results for all time horizons stay well above the watershed 50%, meaning that the algorithm mas been able to perform with an impressive consistency. The wide range of stork predictions it delivers makes it easily usable for a wide variety of investment strategies, from sector rotation to investing into a specific industry.