AI in Finance: How Good is Morgan Stanley’s Foray into Deep Learning?

This article on Deep Learning in Finance was written by Jireh Tan, a Financial Analyst at I Know First.


  • Morgan Stanley used artificial intelligence to study thousands of financial reports made by its analysts.
  • The intended outcome was to determine the confidence of the investment bank’s analysts with regards to different stocks.
  • The strategy proved to be immensely successful, with backtests showing positive returns even in 2018 despite the poor performance of the market during that period.

Gone are the days of the “AI winter”, when artificial intelligence was regarded as a dying field. With the drastic improvement in computing power and the widespread availability of open source artificial intelligence-related algorithms, companies and countries alike are investing substantial amounts of funding into artificial intelligence technologies such as deep learning to ensure that they remain competitive and ahead of their competitors.

Morgan Stanley, the renowned multinational American investment bank and financial services company, is no exception. On 27 June 2019, a report from CNBC announced how the bank used deep learning to study its own analysts to determine their analysts’ confidence in different stocks. 

What is Deep Learning?

Deep learning is a machine-learning technology designed after the human brain. Just as how human beings learn to make sense of the world through knowledge and experience, customized deep learning algorithms also require substantial amounts of training. Deep learning training replicates the many interconnecting neurons in the brain, where the artificial neural network passes input data from the input layer to the many intermediate layers, until it finally arrives at the output layer. By comparing the output result with the actual result, the algorithm can then adjust the relative importance (i.e. weights) accorded to each layer to improve its accuracy in subsequent training rounds. Over an extended period of time, as it gains more “experience”, the algorithm gets better at delivering the expected output based on the input data. Once its accuracy is sufficiently satisfactory, the algorithm can then be applied to new sets of data to gain new insights, ideally ahead of time to improve the outcome of decision-making processes.

Like the human mind, deep learning involves processing data through many layers to draw conclusions on the input information. (Source: Pixabay)

Deep Learning at Morgan Stanley: A Summary

At Morgan Stanley, natural language processing is utilized with deep learning to determine the sentiments of their analysts on different stocks based on their reports. Based on the report, natural language processing at Morgan Stanley involves taking in entire sentences as input, and assigning a score to each sentence based on the confidence that the algorithm believes the analyst to have. As seen in the article,

“Phrases like “great things,”, “definitely a plus” and “much better” can be read and loosely understood as nearly always positive by a deep learning platform.”

By extension, we can conclude that sentences which urge investors to be more cautious will be assigned a relatively lower, or even negative confidence score. By considering the net score of each report, and the correlation between the actual stock’s performance and the numerical value of the analysts’ confidence in this said stock, the algorithm would thus be able to identify the stocks which are more likely to perform better amongst the thousands of other stocks studied by Morgan Stanley’s analysts.

Deep Learning at Morgan Stanley: How Effective is it?

With price movements in the stock market appearing seemingly random, Morgan Stanley’s deep learning algorithm can be tremendously useful in determining the best stocks to trade or invest in. Considering that the algorithm builds upon work done by their analysts, the numerical value it generates can serve as an empirical means of check-and-balance for the bank. This usefulness can range from evaluating the risk-reward ratio of specific investments or trades, to sieving out the best analysts at the bank. In essence, Morgan Stanley’s deep learning algorithm is a combination of its best analysts. Considering how the algorithm improves with experience, it is very likely that insights produced will only get better over time, thus making it a force to be reckoned with

Artificial Intelligence
While the accuracy of Morgan Stanley’s deep learning algorithm will likely improve over time, one must note the many limitations to its practical usage as well. (Source: Pixabay)

However, we also note that the performance of Morgan Stanley’s deep-learning algorithm is limited to the quality of the reports produced by its analysts. Poor and inaccurate reports may have detrimental consequences on the algorithm, which will inevitably affect its accuracy to produce reliable forecasts, both in the present as well as in the future. Furthermore, the fact that the algorithm’s output is based on sentiments obtained from reports may imply that its reliability is less than desirable, due to the fact that it is heavily dependent on human sentiment which it may firstly interpret incorrectly, or which may simply be based on emotions rather than on actual hard facts.

The I Know First Algorithm Utilizing Deep Learning

Here at I Know First, our algorithm is purely based on empirical data and not on any human derived assumptions. Upon building the mathematical framework and presenting to the system the starting set of inputs and outputs, the development of the algorithm is solely based on our proprietary technology. 

Predictive models are constantly being self-developed and tested based on years of market data and historical success of the algorithm’s models’ pool. These models are then validated on the most recent data to identify and improve the most optimal predictive models. Should any inputs not improve the existing model, these are discarded and substituted with other inputs to improve its forecasting accuracy. The resulting algorithm is thus constantly evolving, as new daily data is added and a better machine-proposed predictive model is found.

Over time, the I Know First algorithm will only get better. In recent weeks, we have seen how our algorithm has identified opportunities for a return of 18.37% in just 7 days, as well as returns of 10.93% in 3 days. We don’t just provide opportunities for the short term either – the algorithm has also forecasted a basket of stocks for a return of over 18% in 3 months and of over 16% in 1 year.

You can find out more about the I Know First algorithm, as well as further reports on the success of the past performance of our algorithm here.

To subscribe today click here.

Please note-for trading decisions use the most recent forecast.