Subscribe to DSC Newsletter

Time series prediction: weekly price of the iPhone 7 on the refurbished market

Hello,

We are trying to predict the weekly price of the iPhone 7 on the refurbished market.

We are using an auto-regressive model, which takes as input the weekly price of an iPhone for the last ten weeks, including the current one, and which products as output the price for the next week.

Feeding a SVR with historical data for the iPhone 7 offers a good R2 score in tests.

The point is that when we add historical data of other phones such as iPhone 6, iPhone 5s to the training data, we obtain a model with a better R2 score in tests.

The model is then able to make predictions for all these phones. The weekly prices of these phones on the refurbished market have similar behaviors.

However, I think that by doing this we are leaving the framework of time series prediction.

Do you think that building such a model is a good idea, as it seems to work better than a more traditional one ?

Thanks in advance,

Xavier

Views: 281

Reply to This

Replies to This Discussion

Wouldn't a regression model be better? Prices would fluctuate between categories eg. wear & tear, refurbished, model etc. and having a regression model would help account for these.

You can then plot an average of the prices of models over time to get a "time series" devaluation.

Thanks for your response. 

We performed a SVR (Support Vector Regression).

Finally we switched to multiple ARIMA models (one for each phone and condition).

Xijian Lim said:

Wouldn't a regression model be better? Prices would fluctuate between categories eg. wear & tear, refurbished, model etc. and having a regression model would help account for these.

You can then plot an average of the prices of models over time to get a "time series" devaluation.

Reply to Discussion

RSS

Follow Us

Videos

  • Add Videos
  • View All

Resources

© 2018   Data Science Central™   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service