Depends on your feature extractor. If you have a feature that measures similarity to previously-seen films, then yes. Otherwise, no. If you only have features measuring what each film’s about, and people like novel films, then you’ll get conservative predictions, but that’s not really the same as learning that novelty is good.
Thanks. Now I’m trying to learn a bit about (exponential) moving averages. I know moving averages are used in machine learning, but I’ve also come across them in stock market investing where they are regarded with derision. Can someone explain what their utility is, and how they can be useful when they aren’t in trading? If my financial knowledge is correct, moving averages only indicative profitable moves when there are linear dependencies between fundamental variables and the stocks price. This is true both empirically and is what most technical analysts assume. However, how do we know that for a particular security, price behavior isn’t better modeled as a random walk. Basically we don’t, and we know for any given stock, it’s a pretty good generalization. However, asset classes as a whole, or particular indices, often go up over the long run. It frustrates me that technical traders strategies just create self-fulfilling prophecies. They’re like uninformed (in the market information sense) speculators who act on the same tea leaves.
I don’t understand the reason for using moving averages unless you have reason to believe, in advance, that it will be a good model of the physical behaviour of that which you’re trying to predict. But then, you wouldn’t be conducting predictive analytics then. Wouldn’t neural networks, otherwise, dominate them? I would imagine that they would stumble upon a moving averages strategy if that was evidently a good model for the phenomenon in question.
And yet the risk metrics associated with the implementation of a simple neural network in Quantopian doesn’t seem that attractive. But what can’t neural networks do? They seem like most perfect learning devices ever!
Depends on your feature extractor. If you have a feature that measures similarity to previously-seen films, then yes. Otherwise, no. If you only have features measuring what each film’s about, and people like novel films, then you’ll get conservative predictions, but that’s not really the same as learning that novelty is good.
Thanks. Now I’m trying to learn a bit about (exponential) moving averages. I know moving averages are used in machine learning, but I’ve also come across them in stock market investing where they are regarded with derision. Can someone explain what their utility is, and how they can be useful when they aren’t in trading? If my financial knowledge is correct, moving averages only indicative profitable moves when there are linear dependencies between fundamental variables and the stocks price. This is true both empirically and is what most technical analysts assume. However, how do we know that for a particular security, price behavior isn’t better modeled as a random walk. Basically we don’t, and we know for any given stock, it’s a pretty good generalization. However, asset classes as a whole, or particular indices, often go up over the long run. It frustrates me that technical traders strategies just create self-fulfilling prophecies. They’re like uninformed (in the market information sense) speculators who act on the same tea leaves.
I don’t understand the reason for using moving averages unless you have reason to believe, in advance, that it will be a good model of the physical behaviour of that which you’re trying to predict. But then, you wouldn’t be conducting predictive analytics then. Wouldn’t neural networks, otherwise, dominate them? I would imagine that they would stumble upon a moving averages strategy if that was evidently a good model for the phenomenon in question.
And yet the risk metrics associated with the implementation of a simple neural network in Quantopian doesn’t seem that attractive. But what can’t neural networks do? They seem like most perfect learning devices ever!