Time series analysis is a branch of applied statistics that deals with observed data collected at different points in time. This analysis uses statistical models to understand and predict trends in a process, often by a set of observations taken from that process. A simple example would be plotting the number of customers who walked into a shop every day for 15 days. Time Series Analysis can be used to predict customers in the future, given what has happened in the past.

A time series model uses observations taken at sequential points in time to understand trends and patterns that occurred or are expected to occur in the future. Time-series models are often very useful to predict future values based on trends that have been observed in the past.

When should you use time series reconstruction instead of other methods for modeling data?

Time series models allow you to account for relationships between past and future observations in a way that simple linear or non-linear forecasting models do not. Time Series Analysis can give you a better understanding of how changes in one time period impact future periods by looking at what has happened before. They allow us to understand, for example, whether prices in the market go up or down over time, and how that impacts demand.

Time series models are also useful for predicting future values based on trends that have been observed in the past. This can be especially important to understand trends within a business (e.g., calculating the number of widgets the company expects to sell next quarter), to make projections about future strategies, or even to predict future demand in the market.

What is Time Series Reconstruction?

Understanding the role of time series analysis is just thing beginning. In a perfect world, a time series analysis would include all of the data points and activities needed to make a predictive model for the future. Like most models, more available data should inform more reliable decisions just as models that lack data will be less reliable.

The role of time series reconstruction is to fill the gaps in data in a reliable and efficient way. For example, imagine a social media platform that reports only a portion of data around engagement. Even worse, that data does not extend beyond a few days or weeks into the past. 

Any resulting models and analysis would be, at best, informed by incomplete data sets.  At worst, and more likely, the resulting models might not even be directionally correct.

Appropriate time series reconstruction fills in the gaps so a more complete data set can inform future decisions.

How Does Time Series Reconstruction Work?

There are many different types of models that can be used for time series analysis including moving average models (like exponential smoothing) and autoregressive integrated moving average, or ARIMA models. These are all examples of models where the dependent variable is a linear function of the previous data points, plus some other variables that are not time-dependent.

The simplest form of a time series model assumes that the entire relationship between future values and past values is captured by just one point in time. This method does not allow for any historical relationships to be reflected in the value of the forecast.

One alternate method is to assume that all lagged values are relevant for predicting future values (e.g., past and present). This method does not allow for any additional factors beyond those reflected in the time series analysis and is not very flexible.

Considerations In Using Time Series Analysis

The chosen time-series model should be appropriate to your problem. If your dataset covers a long period of time, you may also need to consider how the events that happen at the beginning of the dataset may affect subsequent observations. This will differ depending on what type of event happened, but it is common for a time series to have a change point after which it is no longer appropriate to use the same model as before.

Time series models are very sensitive to outliers, so if your dataset has any unusual observations that do not represent typical values you will want to remove those points or otherwise account for them in your analysis. For example, if there were time periods when you were not operating normally, it may be appropriate to use a model that accounts for anomalies.

Leveraging Time Series in Your Own Marketing 

Time series reconstruction takes known data sets to create models around existing incomplete information. The aforementioned social media data set, for example, can be augmented and enhanced. The result? More accurate time series analysis across multiple channels, platforms, and campaigns.

The best way to chart the future is with a complete understanding of the past. Time series reconstruction and analysis deliver that understanding. To better understand how Alembic’s Marketing Conversion and Event Correlation engine uses time series reconstruction to help advanced marketing teams make better decisions, contact us here