Sales Forecast Demo

North American freight rail car deliveries (in units) from 1988 through the third quarter of 2007 are as follows:


YEARQ4Q3Q2Q1
2007 150321614317148
200617927190081946618542
200517975169871791415781
200414419117901007110012
20039953825173656614
20024801492541553855
200170207188898211070
200011993127821417916867
199917762163851888221560
199820242183991929017773
199715444120441141411494
199614317148811412914690
199514476140361570816633
199414034148951355610881
19939908828583458145
19927630750959624660
19915943633160036397
19907541765979718892
19897672682178387286
19887864560545984457

From looking at this data, can you see any kind of annual pattern?  That is, are deliveries of railcars cyclical - higher at certain parts of the year, lower at other parts?  Is there any sort of pattern across years, say, up and down every five years?  Regardless of patterns, is there any general trend going up, down, or flat? 

Blindly attempting a quantitative forecast without having a feel for these sorts of issues is pointless.  You should look at a visual plot of historical sales, and then you need to assess two factors: patterns and trends.  For any patterns and trends that you see, what time period is relevant in using these to make a forecast?  At what point does older data lose its value in assisting you to make a forecast? Importantly, what factors in the external environment have a bearing on a forecast?

Click on the following links to see interactive spreadsheets and plots:

[These will only work with MS Excel.  I created these with the latest version and don't yet know how they will work in older versions.  Use the school computer lab if you experience any troubles.]

On each spreadsheet, you will see a slider.  Move it to change the value of n on a moving average forecast and to change the value of alpha on an exponential smoothing forecast.  Each of these smoothes the data, useful in helping to find a pattern when the data is "noisy".