Let’s Talk Time Series: Part 1(Goal Definition)

This is first part of my series on time series forecasting. I would be talking about every aspect of time series forecasting in detail in each part. Before beginning any task we need to have a clear idea of what our goal is and what we want to achieve.

In this article I would be giving the idea about what goes into planning a task for time series and give a brief on how to solve the challenges arise.

In the later articles I would talk about more technical terms. This would be an overview on what time series forecasting is all about.

Time Series analysis vs Time series forecasting

Time series forecasting and analysis are very closely related terms but there is a difference. The time series analysis refers to using the current data available to provide insights on the trends and seasonality in the data . This helps in making important business decisions. In certain cases the time series analysis can use the values from the future to provide better insights for the value today too.

On the other hand the time series forecasting refers to using the existing data and using it to predict the values for the future.

Since the goal of the tasks is different in both these cases the methods employed to achieve them also differ. I would be focusing majorly on the time series forecasting, however some of the techniques are similar in both the tasks.

Introduction

Time series forecasting is unlike any other predictive analysis technique. This is because there is an element of time in it. The value at any certain time is dependent on the previous values. Hence keeping the tracking of previous values is also important.

Considering the diagram today we are dealing with the first part of defining goal. But this is a crucial task since this very task is going to act as a guide and constraint for all the tasks later.

Goal definition?

Before beginning any project or task we need to have a clear idea on what we want to achieve. This is important while dealing with time series forecasting too.

So what questions need to remembered before beginning:

1. Do we need to do time series analysis or forecasting
2. Who would be using these forecast?
3. How far in the future we need to forecast. This is important because the when we go in the future the probability of the context being changed gets higher.
4. Is the process ongoing or you need to predict once.
5. How recent are the data available at the time of prediction?Timeliness of data collection and transfer directly affect the fore-cast horizon.

Different answers will lead to different choices of data, forecasting methods, and evaluation schemes. Hence, these questions must be considered already at the goal definition stage.

Referring to point 3 the forecasting task can be an ongoing task or a single project that needs to be completed for a given time frame. For example predicting the weather is an ongoing procedure where we constantly predict the weather and the task is ongoing for any number of days. On the contrary predicting the number of cases of COVID-19 virus for the next month can be a single task.

Roll-forward forecasting

As I mentioned earlier that it is important to know about how far in the future we want to predict the values. The farther we go into the future the more the uncertainty increases.

Let’s say it is the January of 2021. The task is to predict the stock price of the next three month. We create a forecast in January to predict the price for the next three months. This forecast can be refreshed in february and again in march. This would make the forecast more robust when predicting the values for april. This refreshing of the forecasts based on new data is called roll-forward forecasting.

Who would be using this forecasting?

Before forecasting it is also important to know who would be viewing the forecasting and how they would like it to be presented. If this forecast is a part of some project pipeline then the forecast could be an input to some other component of the project or if it needs to be presented to the end user then it needs to be presented differently.

All these points needs to be considered before beginning the tasks.

Conclusion

Some might think that these intricate details might not play a major role when creating the forecast but a proper goal planning in the beginning save a lot of time later when we are trying to find the appropriate data, methods, and evaluation schemes which we would be discussing in the future articles.

Thanks :)