@drl21, Being better than analysts is one thing and being correct most of the time is another. What I'm saying is that there's typically a fairly large error with GLS prediction, not that they're worse than analyst predictions.
When I say delisting a game is unfair, I mean that delisting assumes that the GLS prediction is correct, though it almost always will have a relatively wide margin of error. The last article I read on the SimExchange quoted an error of 16%, which is hardly what I'd call correct. It's more of a ballpark estimate like most estimates of this kind.
As an aside, what are you basing your assessment of the SimExchange's accuracy on? You're saying that we're pretty close 90% time then what is standard deviation of the error assuming the GLS prediction error roughly follows a normal distribution?
When we're talking about inference, we typically want to know with at least 95% confidence that a particular hypothesis is correct. Say our hypothesis is that the SimExchange's GLS predictions within 1 year are within 10% of the actual final value, can you prove to me that this is true with 95% confidence?
Getting back to the topic at hand, predictive markets are useful when no better method exists. The most common critique is that there is a tendency for those participating in predictive markets to abstract their predictions. Instead of trying to predict say what the GLS will end up being, people will try to predict what the average prediction of the GLS will be -- or in essence people will attempt to game the market. That's why awarding correct predictions is important.
Saying that predictive markets are a decent way of predicting the GLS of a game is fine. But claiming that this same prediction should be how we measure the accuracy of our predictions is just flat out wrong. Delisting a GLS prediction before the GLS is known with a high degree of confidence is tandem mount to awarding speculative investment, attempting the predict the prediction instead of the actual GLS.
That being said, the methodology behind the shorter term measures seems fine since they're awarding DKP on the basis of an actual measure.
2
Being better than analysts is one thing and being correct most of the time is another. What I'm saying is that there's typically a fairly large error with GLS prediction, not that they're worse than analyst predictions.
When I say delisting a game is unfair, I mean that delisting assumes that the GLS prediction is correct, though it almost always will have a relatively wide margin of error. The last article I read on the SimExchange quoted an error of 16%, which is hardly what I'd call correct. It's more of a ballpark estimate like most estimates of this kind.
As an aside, what are you basing your assessment of the SimExchange's accuracy on? You're saying that we're pretty close 90% time then what is standard deviation of the error assuming the GLS prediction error roughly follows a normal distribution?
When we're talking about inference, we typically want to know with at least 95% confidence that a particular hypothesis is correct. Say our hypothesis is that the SimExchange's GLS predictions within 1 year are within 10% of the actual final value, can you prove to me that this is true with 95% confidence?
Getting back to the topic at hand, predictive markets are useful when no better method exists. The most common critique is that there is a tendency for those participating in predictive markets to abstract their predictions. Instead of trying to predict say what the GLS will end up being, people will try to predict what the average prediction of the GLS will be -- or in essence people will attempt to game the market. That's why awarding correct predictions is important.
Saying that predictive markets are a decent way of predicting the GLS of a game is fine. But claiming that this same prediction should be how we measure the accuracy of our predictions is just flat out wrong. Delisting a GLS prediction before the GLS is known with a high degree of confidence is tandem mount to awarding speculative investment, attempting the predict the prediction instead of the actual GLS.
That being said, the methodology behind the shorter term measures seems fine since they're awarding DKP on the basis of an actual measure.