Overview
A time series is a series of measured values with a timestamp attached to each measurement. In most time series analysis, additional calculations and transformations are applied. This process of computing additional features of a time series falls under the category of feature extraction.
Time Series Transformations
- Processing Trends and Non-Stationarity
- De-Trending - when a time series contains a deterministic trend, it needs to be de-trended in order to make it stationarity (see trends)
- Differences : One of the common features that are calculated from a time series is a set of differences. A difference is a number that is calculated from separate data points in the series, typically adjacent points. It can be a simple subtraction of two points, but can also include such calculations as an arithmetic return. Differences are important in dealing with issues surrounding stationarity
- Volatility : is a technique used to measure the amount of variability in a time series, typically in financial data.
- Filters : A filter is a transformation that takes a signal (time series) and produces a new signal. In time series, a filter is sometimes used to compute a smoothed representation of the time series. The smoothed series is often used to represent the trend of the series. (In this case, it would be thought of as a deterministic trend)
- Normalization - feature processing technique that scales features by their standard deviation or some other measure of spread in order to get them on an apples to apples basis
- Technical Analysis - set of features used prominently in asset trading
Tools and Implementation
- Indicators App - an app designed to help users compute various features on a time series without having to engage in scripting.