User:IssaRice/Moral public goods example: Difference between revisions

From Machinelearning
No edit summary
No edit summary
Line 9: Line 9:
So to solve for the highest tax that the nobles are willing to pay, we want to find <math>a > 0</math> such that <math display=inline>u(x_n, x_n/\alpha) = u(x_n - a, x_n/\alpha + \frac{N_n}{N_p}a)</math>.
So to solve for the highest tax that the nobles are willing to pay, we want to find <math>a > 0</math> such that <math display=inline>u(x_n, x_n/\alpha) = u(x_n - a, x_n/\alpha + \frac{N_n}{N_p}a)</math>.


This is <math display=inline>\log x_n + \log(x_n/\alpha) = \log(x_n - a) + \log(x_n/\alpha + \frac{N_n}{N_p}a)</math> which reduces to the following quadratic in <math>a</math>: <math display=inline>x_n(x_n/\alpha) = (x_n - a) (x_n/\alpha + \frac{N_n}{N_p}a)</math>.
This is <math display=inline>\log x_n + \log(x_n/\alpha) = \log(x_n - a) + \log(x_n/\alpha + \frac{N_n}{N_p}a)</math> which reduces to <math display=inline>x_n(x_n/\alpha) = (x_n - a) (x_n/\alpha + \frac{N_n}{N_p}a)</math>, which is quadratic in <math>a</math>. Using the quadratic formula and simplifying, the variable <math>x_n</math> actually drops out, and we get <math display=inline>a = 1 - \frac1\alpha\cdot\frac{N_p}{N_n}</math> as the non-zero solution (the other solution is always zero).

Revision as of 19:56, 26 January 2020

working out the general optimal tax for the example given in https://www.greaterwrong.com/posts/pqKwra9rRYYMvySHc/moral-public-goods

each noble has utility function u(xn,xp)=logxn+logxp, where xn is the noble's own wealth and xp is the average wealth of a peasant (since the donation/tax is distributed equally among peasants, this is the same as saying the wealth of a single peasant).

a tax of amount a shifts the current utility amount u(xn,xp) to u(xna,xp+NnNpa), where Nn is the number of nobles and Np is the number of peasants. This is because each noble loses a, so there is Nna to distribute, and this is divided by the number of peasants.

If each noble starts out with α times as much wealth as the average peasant, we have xn=αxp.

So to solve for the highest tax that the nobles are willing to pay, we want to find a>0 such that u(xn,xn/α)=u(xna,xn/α+NnNpa).

This is logxn+log(xn/α)=log(xna)+log(xn/α+NnNpa) which reduces to xn(xn/α)=(xna)(xn/α+NnNpa), which is quadratic in a. Using the quadratic formula and simplifying, the variable xn actually drops out, and we get a=11αNpNn as the non-zero solution (the other solution is always zero).