Our Auto-Trading Pipeline

5 min readMar 19, 2023


How Predicto generates & places trades daily

Disclaimer: Every forecast/trade seen in Predicto is AI-generated. Our goal is to identify trends/patterns from the past that might repeat in the future. It’s very hard to predict the future, but we can certainly gain insights from the past.

Financial markets are complex systems with countless variables that can impact asset prices. At Predicto, we recognize the importance of utilizing unique datasets and cutting-edge research techniques to analyze and predict market trends. Our auto-trading pipeline leverages the latest advancements in data science to collect and process a wide range of datasets from sources such as Nasdaq Data Link. By training models on this data, we generate daily forecasts for more than 100 stocks, which are then used to identify potential trade candidates. Through continuous research and experimentation with unique datasets, we strive to improve our pipeline’s accuracy and ultimately provide our clients with the most effective investment strategies.

So, if you feel ready, let’s go over each step of our pipeline.

1/ Data collection

At Predicto, we know that high-quality data is the foundation of successful AI systems in financial markets. To help us analyze and predict market trends, we invest in several datasets that we believe have the potential to inform our models. However, we understand that good datasets can be expensive, and we carefully consider our options to ensure that we make optimal choices that align with our research objectives.

We take a focused approach to data collection, selecting those datasets that offer the greatest potential to enhance the accuracy of our pipeline. We recognize the importance of continued experimentation and research to improve our approach, and we are always on the lookout for new and unique datasets to help us stay ahead of the curve.

2/ Data experimentation & model training

We believe that continuous experimentation and research are key to improving the accuracy of our forecasts. While we primarily focus on options-related datasets, we also explore alternative data sources from time to time to identify new patterns and trends. To support this process, we have developed a no-code/low-code platform that enables us to experiment and iterate quickly.

We maintain a separate model for each stock that we track to ensure that we capture the unique characteristics and behavior of each asset. This approach allows us to provide tailored forecasts for more than 100 stocks and identify potential trade candidates.

More details on this part can be found in Scaling Deep Learning Multivariate Forecasting with no&low code.

Our no-code forecasting & experimentation platform UI.

3/ Forecasts & trades generation

We generate daily forecasts for more than 100 stocks using our trained models. Based on these forecasts, our auto-trading pipeline creates a BUYor SELL trade recommendation for each stock, along with a stop-loss, target price, and exit date

To give you an idea of what this process looks like, here’s an example of a couple of trades generated based on our forecasts. This process is repeated for all the stocks that we track, resulting in more than 100 trade candidates per day.

A couple of trades generated based on forecasts.

4/ Trade filtering & placement

Before placing any trade, a filtering process takes place. For each stock, we calculate some metrics based on the recent past. For example:

  • What was the ROI of this model’s trades in the last few weeks?
  • How was the accuracy of this model’s recent forecasts?

We obviously don’t expect all models to perform well at all times, but we want to be able to tell which models have identified or caught some kind of pattern. All our models are ranked based on those metrics.

Ranked models based on recent performance.

We then decide which trades to place based on some additional filters (forecasted ROI, buy/sell side and others).

Trades daily promotion.

Trade placement takes place daily on market open. For our auto-trading, we use Alpaca and we submit bracket orders, which means we provide stop loss and target prices that exit position once a specific price is hit. We also set a “trade horizon” of a maximum of around 10 days.

5/ Performance monitoring

Last but not least, we maintain a monitoring dashboard where we can always see how our trading AI is performing and aswer questions like:

  • How often it goes short vs how often it goes long?
  • What is the performance vs index (SPY, QQQ)?
  • What are the daily entries/exits and P/L%?

Monitoring of a live trading system is really important and allows us to tweak parameters at any part of the pipeline.


That was a brief overview of our pipeline that only touched the surface.

At Predicto, we understand the complexity of financial markets and utilize unique datasets and cutting-edge research techniques to generate daily forecasts for more than 100 stocks.

Our no-code/low-code platform allows us to experiment and move fast to continuously improve our pipeline’s accuracy. By analyzing past patterns and trends, we carefully select trades that have the highest likelihood of success.

We are committed to keep improving our auto-trading service that utilizes the latest advancements in data science to help everyone make informed investment decisions.

Thank you for reading!

— Predicto Team




Stock & Cryptocurrency Forecasting AI. Based on Options Data. Powered by Intelligible Deep Learning models. https://predic.to