Ah sweden, of all my travels, it's by far my favorite country. Especially outside Stockholm. Swedes always think im crazy when I tell them that though.
In fact I was very close to going to grad school at either university of Stockholm or Uppsala.
Slimebeast said:
Ok, I think I understand, but an example wouldn't have hurt. As om mathematics I've completed upper secondary technical school and I have one university course in statistics (just one term, in Sweden btw).
Anyway, so far you're using terms that are way above my level *(plus there's the language barrier).
response:
I realize I have trouble with explanations. At IBM one of my evaluations went like this "Technically Outstanding,has difficulty summarizing to non-technical personel" this is part of the reason I am so eager to help you understand what I am saying. I need practice.
Say you have two models f1(t) and f2(t) where t is time both of which are always increasing. One which usually performs well in the short term, but not in the long term (for example one that relies heavily on only very recent sales data trends). The second uses a larger swath of historical data and while not good at detecting blips is more accurate at predicting the total sales in the end (it maybe difficult to predict when price drops occur, but less difficult to predict what final prices will be). A simple ensemble of the two would weight the first model more earlier and the second more later. If the first tended to predict higher rate of increase(derivative) in the install base than the second then as the weights started to favor the second. There is no guarentee that the weighted average μ(t)=w1(t)f1(t)+w2(t)f2(t) is monotonic, that is that μt(t) >= 0 for all t. If you've taken calculus you know it's going to depend on the weight functions.
This particular example is analagous to why climatology is much easier than meterology a distrinction which seems to be responsible for many of peoples doubts about climate change models.
Slimebeast said:
BTW, the graph interpreted by many as "negative Wii sales" is not a big thing around here (on VGC). Most people thought it was an error and left it at that.
response:
Many used it to discredit the forcast out of hand.
Slimebeast said:
Of course we know that the analysts use intricate statistical models and factors such as shifting demographics and fluctuating economis to make their forecasts. But is that an excuse for their terrible outcomes? BTW, the article we are discussing uses arguments that we are able to understand and judge, and form an opinion about. If the experts are using words like "casuals", "strong brand" and "fad" we're automatically invited to the discussion and analysis, even if we're laymen here.
response:
I am not saying they are not pulling these forcasts out of their asses. I have seen it happen before. It's a particular vice of MBA's and economists. The World bank has had some rather embarassing situations in which in their forcasts of economic growth if only their prefered economic policies were implimented. They had forgotten to change the name of the country on all the pages they had previously advised.
However, if they are using forcasting methods which don't rely on a static model. It's really up to interpretation what the genereted model actually means. And they are going to have to distill these things into things people are going to understand. Wii sales have been abnormally high for the industry. Historically that means fad. The models probably express that. "casuals" means the models are saying things about certain demographic groups otherwise refered to as women. "Strong brand" means nostalgia and the models are probably showing skewed buying parterns of certain age groups in different countries depending what was popular in that place during the groups formative years. Noone who has a job has yet to grow up playing xbox games hence their lack of strong brand. Mario is an objectively annoying character, but I grew up with him so he gets a break.
This all assumes the people talking at these panels are the ones actually developing the forcasts.
