I recently found something that may be of interest to LW readers:
This post at the Lifeboat Foundation blog announces two tools for testing your “Risk Intelligence”:
The Risk Intelligence Game, which consists of fifty statements about science, history, geography, and so on, and your task is to say how likely you think it is that each of these statements is true. Then it calculates your risk intelligence quotient (RQ) on the basis of your estimates.
The Prediction Game, which provides you with a bunch of statements, and your task is to say how likely you think it is that each one is true. The difference is that these statements refer not to known facts, but to future events. Unlike the first test, nobody knows whether these statements are true or false yet. For most of them, we won’t know until the end of the year 2010.
I did not check the test in detail, but I somehow question the validity of the test: As presented in their summary, would not just total risk aversion give you a perfect score? 50% on everything, except for the 0 and 100 entries (where 0 is something like “hey, I do play an instrument, and I know this is total crap, except if I would now be hallucinating, in which case...”). It seems like a test which is too easy to play.
I remember seeing an LW post about why it’s cheating to always guess 50%, but I haven’t found the link to that post yet… I think the basic idea was that you could technically be perfectly calibrated by always guessing 50%, but that’s like always claiming that you don’t know anything at all. It also means that you’re never updating your probabilities. It also makes you easily exploitable, since you’ll always assume that your probability of winning any gamble is 50%. Oh, and then there are the times when you’ll give different probabilities for the same event, if the question is worded in different ways.
It also makes you easily exploitable, since you’ll always assume that your probability of winning any gamble is 50%.
Your probability of winning any two-sided bet is 50%, as long as you pick which side of the bet you take at random. A “rational ignoramus” who always had minimum confidence wouldn’t accept any arrangement where the opponent got to pick which side of the bet to take.
Oh, and then there are the times when you’ll give different probabilities for the same event, if the question is worded in different ways.
That implies a very easy Dutch-book:
Create a lottery with three possible outcomes (a), (b), and (c) - for example, (a) 1, (b) 2, 3, or 4, and (c) 5 or 6 on a six-sided die. (Note that the probabilities are not equal—I have no need of that stipulation.)
Ask “what are the odds that (a) will happen?” In response to the proposed even-odds, bet against (a).
Ask “what are the odds that (b) will happen?” In response to the proposed even-odds, bet against (b).
Ask “what are the odds that (c) will happen?” In response to the proposed even-odds, bet against (c).
Collect on two bets out of three, regardless of outcome.
I recently found something that may be of interest to LW readers:
This post at the Lifeboat Foundation blog announces two tools for testing your “Risk Intelligence”:
The Risk Intelligence Game, which consists of fifty statements about science, history, geography, and so on, and your task is to say how likely you think it is that each of these statements is true. Then it calculates your risk intelligence quotient (RQ) on the basis of your estimates.
The Prediction Game, which provides you with a bunch of statements, and your task is to say how likely you think it is that each one is true. The difference is that these statements refer not to known facts, but to future events. Unlike the first test, nobody knows whether these statements are true or false yet. For most of them, we won’t know until the end of the year 2010.
An annoying thing about the RQ test (rot13′d):
Jura V gbbx gur ED grfg gurer jnf n flfgrzngvp ovnf gbjneqf jung jbhyq pbzzbayl or pnyyrq vagrerfgvat snpgf orvat zber cebonoyr naq zhaqnar/obevat snpgf orvat yrff cebonoyr. fgrira0461 nyfb abgvprq guvf. Guvf jnf nobhg 1 zbagu ntb. ebg13′q fb nf abg gb shegure ovnf crbcyrf’ erfhygf.
I did not check the test in detail, but I somehow question the validity of the test: As presented in their summary, would not just total risk aversion give you a perfect score? 50% on everything, except for the 0 and 100 entries (where 0 is something like “hey, I do play an instrument, and I know this is total crap, except if I would now be hallucinating, in which case...”). It seems like a test which is too easy to play.
I remember seeing an LW post about why it’s cheating to always guess 50%, but I haven’t found the link to that post yet… I think the basic idea was that you could technically be perfectly calibrated by always guessing 50%, but that’s like always claiming that you don’t know anything at all. It also means that you’re never updating your probabilities. It also makes you easily exploitable, since you’ll always assume that your probability of winning any gamble is 50%. Oh, and then there are the times when you’ll give different probabilities for the same event, if the question is worded in different ways.
Your probability of winning any two-sided bet is 50%, as long as you pick which side of the bet you take at random. A “rational ignoramus” who always had minimum confidence wouldn’t accept any arrangement where the opponent got to pick which side of the bet to take.
Please note that I explicitly referred to the test, not to reality.
That implies a very easy Dutch-book:
Create a lottery with three possible outcomes (a), (b), and (c) - for example, (a) 1, (b) 2, 3, or 4, and (c) 5 or 6 on a six-sided die. (Note that the probabilities are not equal—I have no need of that stipulation.)
Ask “what are the odds that (a) will happen?” In response to the proposed even-odds, bet against (a).
Ask “what are the odds that (b) will happen?” In response to the proposed even-odds, bet against (b).
Ask “what are the odds that (c) will happen?” In response to the proposed even-odds, bet against (c).
Collect on two bets out of three, regardless of outcome.