Hyperbolic discounting is an example of intransitive preferences. Yes, real people have intransitive preferences. You shouldn’t model people as utility maximizers. But it’s still an incoherent to the point that it can only be called an error.
We might have a paradigm issue here, but I’d say bite the bullet and accept hyperbolic discounting. Lack of transitivity is just an artifact, and not a problem in the real world. There is an essential difference between cases where you can change your mind and cases where you cannot.
Here’s a simple example. I’m claiming this is extremely typical, and scenarios under which exponential discounting arises are very much not.
When you lend some money to someone for 30 days you should charge interest at rate P. Due to some combination of opportunity cost and risk of the borrower running away with your money or dying and not being able to pay back etc.
When you lend for 60 days instead, the risk of bad things happening between days 31 and 60 is much less than between days 1 and 30. If the borrower wanted to run away, he would. If he survived the first 30 days, it’s a proof that his lifestyle is probably not that dangerous, so he’s more likely to survive the next 30. This decreases the rate to P’ < P. (the same applied to opportunity costs before modern economy). Exactly as hyperbolic discounting predicts.
When you lend to one person for 30 days, and another for next 30 days, your proper interest rate due to risk is back to P. But this is completely different situation, and much less likely to be relevant. And in any case systemic risks of staying in business get lower with time, what should gradually reduce your P.
Usually the points of time where you can “change your mind” correspond to events which introduce all the new kinds of risks and transaction costs and are not neutral.
I’ll understand if due to paradigm mismatch you will have hostile reaction towards this.
Your example doesn’t involve discounting at all. Your nonlinearities are in the probabilities, not the payoffs.
Hyperbolic discounting says that you’re willing to plan to give someone a loan at rate P starting a month from now, but when the time arrives you change your mind about what the fair rate is. Not change your mind in response to new evidence about the probability of defaulting, but predictably change your mind in the absence of any new arguments, just because event X a month from now has different utility to you than event X now.
Of course hyperbolic discounting is a useful heuristic. The paradigm I subscribe to is not just bias. We have these heuristics because they’re the best that evolution or our developing minds could do. That is, they’re pretty good in some other environment (ancestral or childhood), which might be very different. You singled out hyperbolic discounting, among all the biases, but it seems to me much more likely to be maladapted to the present than the other standard biases.
Most of your comment argues that it’s a good heuristic, but your first paragraph (“bite the bullet and accept hyperbolic discounting”) seems to make a stronger claim.
Usually the points of time where you can “change your mind” correspond to events which introduce all the new kinds of risks and transaction costs and are not neutral.
That is a different heuristic than I would call hyperbolic discounting. You can certainly produce situations in the lab where people apply a worse heuristic than that. I expect the two heuristics were more similar in EEA than today.
If you want to make a quick decision, go with your gut and trust hyperbolic discounting. But trust it for a decision on an action, not its intermediate output of utility. It mixes up probability and utility. “If you’re building complex interconnected structures of beliefs” then you have to separate the two and you can’t trust your gut model of yourself because of hyperbolic discounting. People screw up long term planning all the time because of hyperbolic discounting.
By biting the bullet I meant using hyperbolic time as the first default approximation, instead of exponential time. I think exponential time is usually much more wrong in practice than hyperbolic time.
Can you give a concrete example of someone screwing up due to hyperbolic accounting in a case where there’s an objective measure of utility to compare the person’s estimates against ?
There are no objective measures of utility. But just about everyone who has failed a diet or exercise schedule could be seen as failing beause of hyperbolic discounting.
I don’t know. What I was referring to was that people’s estimates of their future utility of some course of action are not constant. And they often vary in such a way that one choice (dieting, exercising, saving...) appear rational when you are planning for it, and when you evaluate it in retrospect, but is unappealing at the time that you actually do it.
Think about our evolutionary history. Presumably, life was less stable, deals less predictable than they are today. In that case it would have been better to have a strong hyperbolic discount rate, while now, when outcomes are increasingly reliable, then that rate should be dropped but it (presumably) hasn’t.
Of course, our intuitive discount rate should never reach the exponential that a model would predict, because there are always new unforeseen factors, but I would contend that the uncertainties have dropped substantially. This would make the particular hyperbolic rate that we intuitively discount payoffs at today biased, while in our evolutionary past it presumably would have been a better approximation of a suitable discount rate.
Hyperbolic discounting is an example of intransitive preferences. Yes, real people have intransitive preferences. You shouldn’t model people as utility maximizers. But it’s still an incoherent to the point that it can only be called an error.
We might have a paradigm issue here, but I’d say bite the bullet and accept hyperbolic discounting. Lack of transitivity is just an artifact, and not a problem in the real world. There is an essential difference between cases where you can change your mind and cases where you cannot.
Here’s a simple example. I’m claiming this is extremely typical, and scenarios under which exponential discounting arises are very much not.
When you lend some money to someone for 30 days you should charge interest at rate P. Due to some combination of opportunity cost and risk of the borrower running away with your money or dying and not being able to pay back etc.
When you lend for 60 days instead, the risk of bad things happening between days 31 and 60 is much less than between days 1 and 30. If the borrower wanted to run away, he would. If he survived the first 30 days, it’s a proof that his lifestyle is probably not that dangerous, so he’s more likely to survive the next 30. This decreases the rate to P’ < P. (the same applied to opportunity costs before modern economy). Exactly as hyperbolic discounting predicts.
When you lend to one person for 30 days, and another for next 30 days, your proper interest rate due to risk is back to P. But this is completely different situation, and much less likely to be relevant. And in any case systemic risks of staying in business get lower with time, what should gradually reduce your P.
Usually the points of time where you can “change your mind” correspond to events which introduce all the new kinds of risks and transaction costs and are not neutral.
I’ll understand if due to paradigm mismatch you will have hostile reaction towards this.
Your example doesn’t involve discounting at all. Your nonlinearities are in the probabilities, not the payoffs.
Hyperbolic discounting says that you’re willing to plan to give someone a loan at rate P starting a month from now, but when the time arrives you change your mind about what the fair rate is. Not change your mind in response to new evidence about the probability of defaulting, but predictably change your mind in the absence of any new arguments, just because event X a month from now has different utility to you than event X now.
Of course hyperbolic discounting is a useful heuristic. The paradigm I subscribe to is not just bias. We have these heuristics because they’re the best that evolution or our developing minds could do. That is, they’re pretty good in some other environment (ancestral or childhood), which might be very different. You singled out hyperbolic discounting, among all the biases, but it seems to me much more likely to be maladapted to the present than the other standard biases.
Most of your comment argues that it’s a good heuristic, but your first paragraph (“bite the bullet and accept hyperbolic discounting”) seems to make a stronger claim.
That is a different heuristic than I would call hyperbolic discounting. You can certainly produce situations in the lab where people apply a worse heuristic than that. I expect the two heuristics were more similar in EEA than today.
If you want to make a quick decision, go with your gut and trust hyperbolic discounting. But trust it for a decision on an action, not its intermediate output of utility. It mixes up probability and utility. “If you’re building complex interconnected structures of beliefs” then you have to separate the two and you can’t trust your gut model of yourself because of hyperbolic discounting. People screw up long term planning all the time because of hyperbolic discounting.
By biting the bullet I meant using hyperbolic time as the first default approximation, instead of exponential time. I think exponential time is usually much more wrong in practice than hyperbolic time.
Can you give a concrete example of someone screwing up due to hyperbolic accounting in a case where there’s an objective measure of utility to compare the person’s estimates against ?
There are no objective measures of utility. But just about everyone who has failed a diet or exercise schedule could be seen as failing beause of hyperbolic discounting.
Without objective measures of utility, what could it even mean to speak of someone’s utility judgements as being biased or wrong ?
I don’t know. What I was referring to was that people’s estimates of their future utility of some course of action are not constant. And they often vary in such a way that one choice (dieting, exercising, saving...) appear rational when you are planning for it, and when you evaluate it in retrospect, but is unappealing at the time that you actually do it.
Think about our evolutionary history. Presumably, life was less stable, deals less predictable than they are today. In that case it would have been better to have a strong hyperbolic discount rate, while now, when outcomes are increasingly reliable, then that rate should be dropped but it (presumably) hasn’t.
Of course, our intuitive discount rate should never reach the exponential that a model would predict, because there are always new unforeseen factors, but I would contend that the uncertainties have dropped substantially. This would make the particular hyperbolic rate that we intuitively discount payoffs at today biased, while in our evolutionary past it presumably would have been a better approximation of a suitable discount rate.