Introduction
Last updated
Last updated
NestQuant's Return and Risk Tournament will operate in rounds and calculate users' points based on the performance of these rounds. All rounds here are market tests. That is, the performance of the model is measured in the future.
Each user is required to submit their model's output, predicting the next 12 hours in the future. Please be aware that we've already adjusted the label for the upcoming 12 hours. This implies that if you possess round data at 7 AM, you can seamlessly merge it with the OPEN_TIME column for direct feature integration and subsequently predict the outcomes for that 7 AM round (which extends until 7 PM; the score for the 7 AM round data will be available after 7 PM). However, in practice, data submission should occur before 7 AM, meaning you're operating at 6 AM and, until this point, only have data from 5 AM. Consequently, if your training strategy involves model submission at 6 AM (with complete data from 5 AM), you'll need to shift the label for the following 2 hours. The same adjustment applies if your strategy involves submitting the model at 5 AM (with complete data from 4 AM), requiring a label shift for the next 3 hours, and so forth.
The results of each round will be displayed when a new round starts.
To facilitate the development of automatic bots, we offer an API that can be accessed through the provided link. This API is specifically designed for Python language, allowing teams to easily interact with our server and scoring system. By referring to the provided link and utilizing our Python-based API, participants can seamlessly create and integrate automated bots, streamlining their interaction with our server and facilitating efficient communication with our scoring system.
At the end of each week, users who do well will have the opportunity to receive rewards corresponding to that week (reward conditions will depend on our benchmark at that time).
There will be two types of submissions:
Input Submission: For this, you will need to provide 3 values - your predicted price, the minimum price you think it could fall to, and the maximum price you think it could reach.
File submission: The second type of submission is a file with two columns - OPEN_TIME and PREDICTION. OPEN_TIME should contain the timestamp for the start of each prediction round. PREDICTION should contain your forecasted label for that round. This file-based submission is better suited for advanced users.
Submission guideline for tournament participants
Go to Model page.
Choose "Backtest now" button for evaluating performance before submitting live predictions, choose "Submit now" if you want to submit forecasts for an upcoming round.
Click the "Add" button if you want to add a new backtest case to evaluate your model. There are two submission types available - choose the option that fits your preferred submission method. You can submit multiple backtest cases to thoroughly test your model before submitting it live. Add as many backtests as you need to feel confident in your model's performance. Once you are satisfied with the backtest results, you can then officially submit your model for live predictions and tournament evaluation. Take advantage of the backtesting feature to refine your model before final submission.
Please be aware that for the second type of submission, which involves input submission, there is a method in place to transform your submitted results into the corresponding prediction. For further details, refer to the Labeling Method.
After submitting your model, please allow some time for the scores to be calculated. You can check the status of your models on the Models page. Click on any submitted model to view details about its performance.
Submissions will be scored against the live target using various evaluation metrics. For this tournament, submissions will be assessed based on two primary scores:
Correlation: The model's significant contribution lies in its correlation, meaning it tends to closely track the same trajectory as the target. This correlation is a crucial aspect that demonstrates the model's effectiveness in capturing the movements and patterns exhibited by the target.
Movement Score: The contribution pattern observed on the Movement Score indicates that the model exhibits a strong predictive capability when it comes to forecasting the direction of price movement. This implies that the model shows proficiency in predicting whether the price will move upwards or downwards, highlighting its ability to anticipate the general trend of price fluctuations.
