Forecasting When There Are Only 2 Known Values in Non-Consecutive Years

You want to model nearly a decade's worth of data based on only two years within that decade and include demand/supply shocks. I am not optimistic about the reliability of this operation.

Standard procedure I would expect to follow.

1) Choose your modeling equations and algorithms. I cannot know from this side of the internet what equations are valid. Depending on what you mean by "constant rate", I could see a straight line y=mx+b type equation or an exponential growth type model y=b*m^x. I have not idea how you would want to handle upsets.
2) Use the known data points to estimate the parameters of your model using an appropriate regression algorithm
3) Compute the values in for the unknown data points.

Reliability seems like it would be quite low, and would depend greatly on the assumptions you choose to use in determining the regression model. If you want to delve deep into the statistics of it all, I'm sure there are ways to estimate the reliability.

I am not sure what "the best" (TM) method is. What you are describing sound to me like a straight line, constant reduction:

Difference = 275-409=-134
difference/5 = -26.8 change per year (slope or m in y=mx+b) -- basically subtract 26.8 from 409 for each year past 2007.
(Note that I find this easier to understand without using percentages -- percentages are too easy to confuse with an exponential growth model)

Usually, straight lines like this are considered "the easiest" models to use. I can't say if "the easiest" is the same as "the best".

Originally Posted by abern079

By "constant rate", I meant that I calculated the % change btwn 2012 and 2007, which is ~ -37% in this case. I then divided that rate by 5 (the number of years btwn 2012 and 2007), which equals ~ -6.5%. So, I basically predicted that the 2007 number would decrease by 6.5% annually, until the 5th year (2012), where total percent change between 2007 and 2012 will, again, equal ~ -37%.

If you want a constant periodic percentage rate of change, you are necessarily describing exponential growth. For the numbers that you have, the correct annual percentage rate of change is:

=(275/409)^(1/5) - 1

formatted as Percentage. To demonstrate, if that calculation is in A1, note that if B1 is 409 and we enter =B1*(1+$A$1) into B2 and copy B2 into B3:B6, B6 displays 275.

On the other hand, if you want a constant periodic amount of change, you are necessarily describing linear growth. For the your numbers, the correct annual amount of change is:

=(409-275)/5

formatted as General or Number or something similar; but not Percentage. To demonstrate, if that calculation is in A2, note that if C1 is 409 and we enter =C1+$A$2 into C2 and copy C2 into C3:C6, C6 displays 275.

Both are equally "good" ways of describing the average annual "rate of change" over some number of years, based on just the beginning and ending valuations.

You just have to decide which model you want. (And there are other possible choices, as well.)