AI stock forecast

Predicting the direction of the stock market is what investors and traders do every day. It involves screening out a lot of irrelevant information and identifying the key aspects that will make each stock rise and fall. Part of this is finding recurring patterns that lead to future price changes. But, of course, once everyone has noticed these patterns, the market adjusts and there is no more profit there to be found. This makes it a race to spot these aspects first and most clearly.

After recognizing that the world is now filled with data and information, much more than anyone could possibly hope to read, we developed MyShare. Our goal is to capture all that information and distil it for our users so they can see the important factors and make better investment decisions.

Why not just use ETF

Another way to invest is just to use ETF and buy a little bit of every share. But if everyone does this, then bad companies and good companies will all be equally rewarded with the same investments. Someone has to decide which companies get funded, and the better the quality of those decisions, the better the economy will be. So we see providing clearer more useful and insightful information as a way to help investors help the economy grow better companies in the future.

What inputs can predict the stock market

The inputs we take for our financial prediction software includes:

* News – this includes blogs, twitter and other news-like information * Company fundamentals (their accounts) which are the most accurate reflection of what companies are actually doing with your money * Commodities like gold, silver, wheat and gas prices, which tells you how much the raw material costs are * Bond rates which tells you how much is the cost of capital (the cost of borrowing money) * Economic information like interest rates, unemployment rates and other national figures

These forms the ingredients to our financial prediction models.

How do we pre-process market data

We take these inputs and normalize them into a standardized input format.

For news, we identify which stocks each news item is mentioning, who they talk about, and what factors they mention inside the stories. When we display news, you will often see a little chevron or arrow beside the news. Chevrons tell you if we thought the story was good or bad for the stock involved, while arrows reflect our opinion on how the news affects the wider economy. Some news (especially about the pandemic) might be good for a health company but bad news for the economy as a whole.

For for economic and financial information, we apply filters to convert them into a form that can be compared from stock to stock and country to country, even of different sizes. For example we might look at the income to sales ratio which is a percentage so doesn’t simply vary according to the size of the company.

Training a stock market prediction algorithm

We then train our models using a multitude of machine learning algorithms, like natural language processing and deep learning. This creates a prediction model from which we can produce forecasts results. We do this training periodically, typically once a week to make new and revised prediction models. Then we use our models on the incoming data to make our final predictions as the information changes.

Testing a real-time AI algorithm

Once we have a baseline set of forecasts, we check each model by using a test-train set split. This means, we only show our algorithm historical data for up to a few years ago, and see how well it would have done on the last few unseen years. This gives us a measure of whether the software is really on to something or was just lucky. Quite often this process shows up a few bugs in the software, or huge dependencies on just one or two factors in the input data and that tells us where the software needs improvement.


Only after checking the model to make sure it is reasonably robust, do we generate new figures for the website. If you happen to visit when there is no data present, it probably means we are in the process of loading up new data.


MyShare is a new site, so there are still lots of areas we need to improve, both in the display and calculation. Occasionally we mis-tag news or get the wrong account information or something. We are continuously improving these things, but if you spot something wrong, please do let us know using the feedback section below.

As the site becomes better, we want to expand into other areas like crypto, options and other investments that people like to buy and sell.