This client gained a substantial competitive edge by equipping their business with the unprecedented intelligence & predictive power of the AI & Analytics Engine to project stock prices for every ASX instrument one year into the future.
The client tasked PI.EXCHANGE with building a predictive model in the AI & Analytics Engine that they could use to quantify the risks present within a given portfolio of stocks, and ultimately use these new insights to help their clients make more informed and higher-performing decisions within their investment strategies.
Given the inherent challenges of quantifying the level of risk in such a volatile environment as a stock exchange, PI.EXCHANGE aims to provide unprecedented insights into ASX data by predicting stock volatility with a model that constructs predictive confidence intervals around stock prices one year from the observation date for every stock on the ASX. This confidence interval will be used to quantify the expected level of return and the probability of achieving that level of return for each instrument.
"We needed an affordable yet powerful AI solution to provide predictive insights into stock volatility. The speed of delivery, the robust deployment stability and their customer services were impressive, and provided us with a competitive edge to help our clients make more informed and higher-performing investment decision."
To first prepare the data, 10 years of ASX stock price data was downloaded for all available instruments. For each time point within the data, the actual stock price one year ahead is created in the time point which is then used as the target in a regression model. From these values, a predictive model is built to construct the desired confidence intervals.
Data wrangling within the AI & Analytics Engine benefits from completely automated & guided feature engineering, including a suite of high-performing time-series functionalities which are used in this case to automatically generate 796 time-series features encapsulating volatility of stocks, the correlation between stocks, and trends within each stock's historical prices and how these trends are correlated across all ASX instruments. This is then packaged into a completely repeatable data-wrangling recipe.
This prepared data is now fed through to the AI & Analytics Engine's unique Model Recommender, which draws upon a rich library of finely-tuned ML models and identifies top performers based on potential financial and time costs - all completely automatically and without any up-front investment. For this use case, the XGBoost model is selected as a high performer and is then used to translate the data prepared on our platform into predictive data for future stock prices. To further manage the volatile and uncertain nature of stock prices, this predictive data is used as the mean (average) of the distribution, and an estimate of the variance of the stock's historical data together with the mean price prediction is utilized to produce a numerical confidence estimate of achieving that level of return.
Finally, this model is deployed seamlessly on our Engine along with the 796 time-series features which are automatically deployed and run on a daily basis, taking advantage of the repeatable data-wrangling recipe generated earlier, and further benefiting from robust deployment stability operating with 0% downtime.