CHECK LIST FOR A3

CHECK LIST FOR A3

Specific forecasting application related to a business case study You need some knowledge of the business.

You are asked to analyse time series data. So, your data will either be time series or data that can be aggregated into time series within BI software. For example, a car dealer would have a record of sales: each sale would be a case; for each case we have make, model, price, and date. BI software can read the file and construct a time series of monthly sales income.

Describe the time series, and original data if different, in terms of variables with unit of measurement, time period, and time step.

Discuss the data quality: missing values, highly improbable outlying values. What have you done about these?

Students may ask how many data points are needed. This is a hard question to answer given the type of forecast. However, I say at least 30 lines would be ideal, if not they will keep asking you the question.

If the data are publicly available, give the data source including the url. If the data are not available, explain why. For example, a company you work for will allow you to use data for your project but are not prepared to make the source data generally available. this would need to be a very short explanation as they have only 1000 words for the whole report.

You may use several time series for your project, which extends the scope of applications, but you do not have to do so. If you have several time series you can investigate Granger causality, or how the variables are associated over time (co-integration).

Applications

Needs to be a business problem/opportunity to take advantage of, where forecasting is needed. Not any other forms of analytics.

There are many possible applications. But, remember that you have to make it into a case study. For example, you work as a financial advisor. Dave, of the recommender system video, wants to invest the proceeds in one Australian bank share. He intends selling in one years time to pay for a holiday on the Barrier Reef. He wants to select just one bank because of the high dealing cost. Which bank would you recommend.

Manufacturing/retail. Are sales equal to demand or can you sell all you produce?. You have

Financial Choosing between bank shares; currency hedging you will receive payment in Erehwon $ when project completed in one years time. Should you purchase an option to exchange Erehwon $ to AU$ at todays exchange rate in one years time (export credit guarantee)?; Granger causality between BHP and Boart Longyear. Remember VAR is a stationary model. Share prices are typically modelled as random walks with drift. So, it may be best to consider log-returns of share prices.

Economic unemployment, base rate.

Environmental OzoneNASA Ozone Watch: Latest status of ozoneWhy is this relevant for your business?.

Public health WHO data, polio vaccination programmes

Transport Road traffic accidents, interventions to reduce, synthetic controls.

Outstanding past projects

All these were based on students current or recent employment and used data from the organisation. They included:

Manufacture and sale of fashion goods

Bank customers and the effect of moving some services on-line rather than in the branch. An overseas location.

Timing of purchase of water rights for Australian farmers growing nut crops.

Share bonus scheme for electronics company. Better for the company to buy shares and keep in an account or to only buy the shares when an employee leaves the schemeMethod

Software includes: Excel, Tableau, Exploratory, Orange

ARIMA (including SARIMA) good for finance. Models random walks with drift. working with log-returns.

H-W adaptive if non optimised. Optimisation tends to give beta = 0 with an initial estimate equal to the average gradient.

Prophet

Orange VAR + Granger Causality

We dont expect a comparison of different forecasting methods. However, it is quick to check RMSE of fit using ARIMA and Prophet, in Exploratory, and you can mention this as one of the reasons for preferring one over the other.

Provide limits of prediction for your forecasts. Note that 0.80 in Exploratory refers to the complement of the upper tail area, and hence 60% Prediction Interval.

Graphics

You need at least one plot showing your data and forecasts. Make sure the plots look reasonable. Clearly label plots and give them figure numbers. Refer to plots in your report.

Conclusion

What was the use of the forecast, and the value of the forecast in your case study? Be specific no marks for a list of all possible uses in general cases.

The length of the report is around 1000 words.