Out of sample data


I am wondering if the out of sample data will be the same set of currency pair in the same layout because we are using some statistical models. So if we train the model using back test data and we don’t want to train the model again but just to use the model to predict and set weight in the out of sample data, will this work ?

We will for sure use the same set of currency pairs but we may stress test your payout on other data / currency pairs. Your logic should be sound and and not dependent / hard coded on any specific currency pairs. Hardcoding of data should be avoided at all costs, it usually results on model that generalize badly and do not allows the Alphien team to do test your submission.

Happy coding ! Lionel.

So in this case we have train the model in the out of sample data as well and it should be part of the payout. Is there any limit on how many days we can look back? In the back test data, we have to generate first weight no later than 2018-01-02 which means our maxim looking back period is 1 year. Is there any similar constraint for out of sample test?

We do not provide any information related to the out of sample test, your objective is to find the best model which generalize well using the data that you have. If your model requires retraining, rolling training, or anything else. Please explain it in your notebook submission. It can indeed be part of your payout.

Is using the column name allowed in the payout function?

Say we just want to trade on XXXUSD currency pairs, we will have to take the column name so our payout function can recognize them.