Q. 127  Time Series Analysis relies on historic data. 

 

When is it of very high relevance? And when should we not rely on Time Series Analysis for forecasting? 

 

Note for website visitors - Two questions are asked every week on this platform. One on Tuesday and the other on Friday.

• All questions so far can be seen here - https://www.benchmarksixsigma.com/forum/lean-six-sigma-business-excellence-questions/
• Please visit the forum home page at https://www.benchmarksixsigma.com/forum/ to respond to the latest question open till the next Tuesday/ Friday evening as per Indian Standard Time.  
• The best answer is always shown at the top among responses and the author finds honorable mention in our Business Excellence dictionary at https://www.benchmarksixsigma.com/forum/business-excellence-dictionary-glossary/ along with the related term.

 

 Time Series is a sequence of measurements of some quantity taken at different time or at equally spaced intervals. Time series models have following components:

1. Long term trend.

2. Cyclical effect 

3. Seasonal Effect 

4. Residual or error effect

 

Time series is of high relevance and has two goals which are given below:

1. Identifying the nature of phenomenon which is represented by the observation  

2. Forecasting means predicting the future values based on time series 

 

in present scenario the world is getting more and more connected, instrumented and smart intelligent and different sensor with latest technologies and other technical support which is collecting thousands of measurements every seconds thus leading to an enormous amount of data being produced on daily basis. In order to get best result from this data we should manipulate in an efficient manner.

Time series analysis plays a very important role. it has so many application out of which some are below:

1. Forecasting: To predict future values for the measure based on the previous data. for example just want to know sales of automotive vehicles in coming financial year based on previous sales data.
2. Cross -Correlation: used to find out how changes on one time series with other time series behavior. for example effect on sales of ice cream, cold drinks as the temperature or summer increases, effect on stock value of two different companies. 
3. Auto - correlation: it is like cross -correlation where an input in a time series correlates it with itself at a different time to discover periodicity. for example accuracy of parts over a period of time at a particular location in fixture or temperature in a given location may be repeated many times in past.
4. Similarity: it is used to measure distance between two or more time series. for example ECG waves chart: we can check if a 5 sec or 10 sec long shape already happened or not in past.
5. Anomaly Detection: it is the ability to discover statistical anomalies automatically in the signal based on past vales.

There is a need to smooth the data if there is any irregularity or any unwanted component found while recording the data from signal. and these unwanted components generally called' noise" , so it is necessary to separate 'noise' from 'signal'. Below is some common techniques to smooth the data:

1. Moving average of size n
2. Central moving average
3. Weighted moving average
4. Exponential moving average

Some times we should not rely on time series analysis for forecasting. This is the case when there are some outlier or any seasonal pulse or any sequential set of outlier with almost same magnitude or same trend. all these are interrelated as A pulse is the difference of a step on the other hand a step is the difference of the time trend. these outliers are not well fitted in the available model. in these cases we must check the source of data that may be faulty or have any error in its accuracy. so if we go time series analysis of a inaccurate date then it might be not useful and we cant rely on its forecasted value so we must have accurate data otherwise we cant rely on time series analysis. we must adjust such points or occurrences as majority of forecast techniques are based on average and any arithmetic average is very sensitive to the outlier value so if there is any outlier it means average is shifted due to its sensitivity so we should not rely on that analysis . 

 

Time Series Analysis is the kind of trend analysis for for data points observed on a regular time interval.
As this technique helps in understanding the past trend of data points / behaviours being observed, we can also predict the forcast of same data points considering all other factors remaining constant / minimum change.
It is of very high relevance when we need to refer past performance and predict / plan our future actions considering no major policy change / external factors change.

Example we need to predict our people capacity i.e. average people present in office on a monthly basis, by monitoring number of leaves on monthly trend for past few years.

Time series analysis can help predict through seasonal variances for a longer time and understand people taking more leaves during DIWALI or End of Calendar year, hence less capacity is predicted on intervals. 

 

If we need to predict number of calls / request in a BPO from clients by reviewing count of calls / requests recieved on a monthly basis.
This cannot be done through Time series analysis as external factors majorly influence month on month behaviour of the data, hence Time series analysis / seasonal analysis will not support or provide better outcome.

Time Series Analysis is a very reliable tool when the factors that influenced the performance of the "metric" under monitoring "remained same" or if we are in a position to quantify the impact of the factors on the Metric with 95%+(or desired) accuracy. In such a scenario "Forecasting" becomes feasible and the Best possible Forecating is Practical !!

 

However, in the real world, New Factors keep cropping up (sometimes even unknown to the Observer) Eg : Retail, ECommerce and ever evolving Consumer Benahvior and Preferences. When Factors Influence alters significantly and when Unknown Factors start impacting the measured "Metric", we will start seeing Forecasting going for a Six and the search for the "Altered Factors/Magnitude of Influence" begings...once again !!

Can we safely say : in a faster evolving Market Reality(particularly in complex environments like Retail & where customer behavior has numerous factors influencing), forecasting based on Historic Data is becoming difficult, by the day !!

Time series analysis is very important as it forecasts based on past facts and performance. It is a whole science, which has proven without any doubt to be very precise. It discounts the regular spikes, shocks and all relevant factors under which a process performs. Many companies rely heavily on this type of analysis to effectively predict their products future.

 

However what Times series analysis lacks is the ability to discount the "game changer events" and "art combination".

 

Example for game changer event is entry of Pitanjali in FMCG and Ayurveda definately would have given nightmares to the giants in these fields and changed the axis of the Time Series analysis for them.

 

Example of Art combination with Science. This is very important as Time series is a pure Science and its combination with Art would make a deadly combination to look beyond the curve at the Event Horizon.

 

For me in my heart I believe that the famous poem/ quote "There is a tide in the affairs of men" very purely and very holy signifies the combination of art and science.

 

History is filled with examples of successes and failures of such cases. Nokia failing in Android, Apple succeeding in maintaining niche in own OS and many more.

 

 

Time Series is a sequence of well defined data points recorded at specific time intervals over a period of time. In Time Series Analysis, while the size of time interval can vary, the interval itself is fixed. For example, if you decide to measure the temperature of a room, you can decide to measure the temperature on an hourly, 12 hours once or daily basis. Once you decide on decide on the frequency (say daily once), you need to follow the pattern of recording temperature on a daily basis over a period of time to arrive at Time Series Analysis.

 

In Time Series Analysis, you analyze historical data points to predict the future. As the name suggests, time is a critical dependency in this analysis.

 

There are 4 key components of Time Series Analysis

1. Trend - which essentially indicates increase or decrease in the series over a period of time over a long period of time
2. Seasonality- short time variation occuring Diego seasonlitt
3. Cyclicity - medium term variation caused by circumstances which repeat in irregular intervals
4. Irregularity - variation due to unpredictable factors which generally are non repetitive.

 

Advantages of Time Series Analysis

It can help in trend analysis, forecasting, intervention Analysis i.e. any changes by introducing some variants and study dependency between two time series data.

 

Limitations of Time Series Analysis. 

We cannot use Time Series Analysis when,

A. Output values remain constant over a period of time.   
B. When output follows a certain pattern like Sinewave. 

Time series will not be of help if we have very few data points (say less than 50) or if the time intervals are not regular. It can also not be used where the environment is so dynamic that it cannot  be uncovered and built  into the model basis the existing data.

 

The best answer is that of Himanshu Singh as this is the only answer that highlights an example of when one may or may not use time-series.

 

Vastupal outlines the details around time series while Manjeet and Prashant mention specifically when it may not be used.