Word “ARIMA” in Tamil language the means **Lion**.

Everybody is curious and anxious enough to know what the future holds? It’s always exciting to know about it. Though there are various forecasting models available in this post we will look at ARIMA. Welcome to the world of Forecasting with ARIMA.

## What is ARIMA?

ARIMA is a forecasting technique. ARIMA– **A**uto **R**egressive **I**ntegrated **M**oving **A**verage the key tool in Time Series Analysis. This link from Penn State University gives good introduction on the time series fundamentals.

## What is the purpose?

To Forecast. The book Forecasting: principles and practice gives a very good understanding to the whole subject. You can read it online.

## What kind of business problems it can solve?

To give examples the following are some of the use cases of ARIMA.

- Forecast revenue

- Forecast whether to buy a new asset or not

- Forecast of currency exchange rates

- Forecast consumption of energy or utilities

## What is mandate to get started?

- It is very important to have clarity on what to forecast. Example if you want to forecast revenue whether it is for a product line, demography, etc., has to be analysed before venturing on to the actual task.

- Period or the horizon in which the forecast is to be done is also crucial. Example: Monthly, Quarterly, Half-yearly etc.,

## What are the preferred pre-requisites on data for Time series forecasting?

### Updated after comment from tomdireill:

- Data should be part of time series. That is data which is observed sequentially over time.

- It can be
**seasonal**. Means it should have highs and lows. As per the notes from Duke University it can be also applied on flat pattern less data too.

- It should have
**trend**of increasing or decreasing

**outliers**

can be handled as outlined here http://www.unc.edu/~jbhill/tsay.pdf

Ok, Now we got to understand what is essential to get started on forecasting, before we devolve lets work on the steps.

**5 Steps towards forecasting:
**

In the next post we will take up an example and work on the above steps one by one. Keep waiting.

Seesiva,

On Points 2,3 and 4 you say the data should be seasonal, should have a trend and shouldn’t have any outliers. Why can’t you use ARIMA on flat patternless with outliers? No dataset would fit these restrictions as these occur all the time.

Tsay laid out how to handle outliers http://www.unc.edu/~jbhill/tsay.pdf

Balke laid out how to handle changes in trend http://www.jstor.org/stable/1391308

Dear Tomdireill,

Thanks for highlighting. Its preferred to have these characteristics for better results in general for forecasting, those points are not specific to ARIMA. Thanks for those links. We can use ARIMA on flat patternless data with outliers, there is a paper on this topic from Duke university http://people.duke.edu/~rnau/Notes_on_nonseasonal_ARIMA_models–Robert_Nau.pdf

Seesiva,

So, we agree…good. I think the pre-requisites you have listed should be changed then.