I've figured out an interesting way to predict end of gen sales, and it involves statistics and listening to analysts.

I started thinking about this starting from a few self evident truths

1) we can't just believe analysts, because they can be very wrong

2) We can't just go based off statistics based on current sales figures because too much can happen to change things in ways that a statistician with no knowledge of gaming could predict

3) Any analyst could, on Dec 30 2012, predict the end of 2012 sales for cosoles with decent accuracy

3.1) Following the above, we can assuming that the errors of analysts will tend to 0 as the generation progresses

If anyone disagrees with these basics, go ahead.

So I graphed the marketshare predictions that various analysts made over time. I only used worldwide predictions, and they all came from 

http://vgsales.wikia.com/wiki/Analyst_predictions

Some predictions are end of gen, some are as early as end of 2010 - lets call it experimental error and move on.

Now, I then put in some trendlines for this data and extrapolated for end of generation, with results shown below

Which gives us final market shares of 26% 360, 23% Ps3 and 38% Wii - total is 87%

However, It seems that a couple of those points are fishy, and could be considered as outliers. The 4th and 5th were removed and this graphed below

Giving us end of gen predictions of about 21% for the 360, 23% for the PS3 and 44% for the Wii. Adds up to 88% total, which isn't bad accuracy (considering the method)

Multiplying each of these figures by 100/88 we get

360 - 23.9

PS3 - 26.1

Wii - 50 exact

Discuss

Also, inb4 people taking this way too seriously :P

I have a serious problem with your methodology...





Wii should have been blue, 360 green, and PS3 red.

Ok, enough with the jokes. Here are a couple of real questions:

1. I assume the Y-axis is percent market share. I suppose the X-axis is time, but what are the units?

2. How did you decide what type of regression to use for each console? Two of the lines look logarithmic, while one looks linear. Although the data may have just been as such to flatten out that logarithmic trend line.

Yeah, I only realised too late that the default colours were put in a bad order, sorry :P

as for my trendlines, I initially started off using high order polynomials, because they had the lowest r^2 value, which for any of us lucky enough to have escaped stats courses is how closely the two sets of data correlate was highest that way. However, that failed miserably when I extrapolated the lines.

As part of my initial assumptions, I stated that I believed the analysts predictions would, as time goes on, tend towards the actual answer. The only trendline that will tend towards a certain value but never reach it (or go past) in excel is exponential for a decay and logarythmic for an increase. The PS3 and 360 both have exponential decay, and the Wii logarythmic increase

And yes, Y axis is marketshare in percent. X axis is time in number of months since 1 month before the first prediction (the 1 month offset was required because without it, excel refuses to use log trendlines - dividing by 0 presumably being the problem)

Alright, now it just looks like you need more data points.

Quick! More analysts!

Yep more data points, and I'm going to guess the reason you didn't include the relevant r^2's is because they weren't good due to your small sample sizes.

I got as many data points as I could - lots of predictions were USA only, or only included sales for 1 or 2 consoles

r^2's were a bit over 0.8 I think - so yeah, not great

I'm just givin' ya crap scottie, it's a fun little exercise and a lot more useful than just sayin "PS3 is CRUSHING 360 and will 4EVA!" or something along those lines.

But for the first 3 years, wasn't every analyst predicting a mirror image of what actually happened? Wouldn't it just be easier to flip over the initial data and assume that the PS3 and 360 are the ones that will get negative sales and the Wii will double and triple its sales in its 6th and 7th years?