4th

Summary

• Rule-based approach without machine learning.

  • Rank return_1day in descending order.
  • If ExpectedDividend is greater than 0, make it the lowest.

• We posted the following discussion.

  • https://www.kaggle.com/competitions/jpx-tokyo-stock-exchange-prediction/discussion/359151

Features Selection / Engineering

SKIP(Because it does not use machine learning)

Training Method(s)

SKIP(Because it does not use machine learning)

Interesting findings

• What was the most important trick you used?
  • Luck!
• What do you think set you apart from others in the competition?
  • The timeseries-API allows models to be evaluated in a way that is close to real-world problems.
• Did you find any interesting relationships in the data that don't fit in the sections above?
  • No, even if we did find them, there is no guarantee that they would be useful for forecasting because market distributions change.

Simple Features and Methods

SKIP(Because it does not use machine learning)

Model Execution Time

• How long does it take to generate predictions using your model? 71.0s

References Discussion

• Could this competition have a lucky winner?, https://www.kaggle.com/competitions/jpx-tokyo-stock-exchange-prediction/discussion/320323
• Shorting dividend days = easy score boost, https://www.kaggle.com/competitions/jpx-tokyo-stock-exchange-prediction/discussion/320836