Amazon is starting a forecasting service for business, using methods employed in its own operations. Users will upload their time series data and use recipes to generate forecasts. The time series recipes Amazon offers (documentation is at the link) are
ARIMA
DeepAR+
Exponential Smoothing (ETS)
Mixture Density Network (MDN)
Multiquantile Recurrent Neural Network (MQRNN)
Non-Parametric Time Series (NPTS)
Prophet
Spline Quantile Forecaster (SQF)
In a sense, what traders are trying to do is forecast returns and position themselves accordingly. I am familiar with ARIMA and exponential smoothing, but some of the other models are new to me. Have people tried these approaches on financial time series? I do understand that asset returns are mostly random, but even a small amount of predictability can be exploited.
ARIMA
DeepAR+
Exponential Smoothing (ETS)
Mixture Density Network (MDN)
Multiquantile Recurrent Neural Network (MQRNN)
Non-Parametric Time Series (NPTS)
Prophet
Spline Quantile Forecaster (SQF)
In a sense, what traders are trying to do is forecast returns and position themselves accordingly. I am familiar with ARIMA and exponential smoothing, but some of the other models are new to me. Have people tried these approaches on financial time series? I do understand that asset returns are mostly random, but even a small amount of predictability can be exploited.
