Greece Debt Relief, Game Theory
Bruce bd Mesquita (BdM) developed a method to predict how decisions are formed within a group of people, nations, alliances, parliaments so forth. It does not matter whether the group has rivalry or consists of generally like-minded actors; the method still works. See links at the bottom of the post for the successes.
The first version of the method took 3 data points per stakeholder to to base further computation on. Per issue, these measurements are 1) clout (how much influence an actor has on the issue), 2) his/her position represented on a 0-100 scale, and 3) a salience measure, how much an actor cares about that issue. Later versions added hawkish / doveness as well.
Then BdM processes this data as part of a repeated game in which he finds the optimum decision point for all parties involved, across all dyads, taking into account whether actors are hawks, doves, passives, etc. At every iteration all actors try to bring someone else to their position, whether successful or not, or how much depends on the 3 attributes shown earlier, and BdM's special algorithm. if someone has high salience on an issue for example, has lots of influence, this person might not be swayed, but can be if some of the base numbers change, etc. Performing this many iterations, a group optimality is reached, and that is the final decision.
A very important feature of the method is that clout, position and salience data points need not be researched. Any observer who reads enough papers, follows politics can come up with them. This is in favor of the method, that its base data is commonplace.
"Can representing positions between 0-100 scale work?". It does surprisingly. In fact I'd suggest anyone who thinks about issues to think of this scale, in a way we all use it in our heads, no matter how complex, if granular enough the issue can be represented as numbers between 0 and 100.
The Mean Voter Position
A very rough version of this computation would be simply computing the mean voter position, just to demonstrate the base data of the method.
For example will the EU give debt foregiveness to Greece during this crisis [in 2015]?
Ran some numbers on Greece/EU negotations, using Predictioneer's Game format,
Stakeholder Clout Position Salience
0 Greece 100 40 100
10 USA 30 60 100
1 IMF 20 70 100
2 Juncker 10 80 100
3 Germany 100 100 80
4 France 70 100 100
5 Dijsselbloem 70 100 40
6 Tusk 70 100 40
7 Italy 50 100 100
8 ECB 20 100 70
9 Spain 20 100 100
mean voter 82 weighted mode 100.0
The positions are,
0 Debt Relief No Reform
30 Debt Relief Half Reform
60 Debt Relief, Full Reform
80 Debt Relief Soon, Full Reform
100 No Debt Relief, Full Reform
I calculate mean voter position 82 and weighted median voter position is 100 from this -- the final outcome being somewhere btw those numbers, which contains reform in all cases, and, very unlikely but, some debt relief in near future. Juncker apparenty dangled this concession in front of Tsipras, debt relief "sometime this year" as a last-minute effort to salvage some deal last week, but was declined (reportedly Juncker was rebuked from the EU side for this action). It is clear T. does not want spending cuts / programs imposed on himself / his government.
def weighted_mode(df):
df['w'] = df.Clout*df.Salience
df['w'] = df['w'] / df['w'].sum()
df['w'] = df['w'].cumsum()
return float(df[df['w']>=0.5].head(1).Position)
def mean(df):
return (df.Clout*df.Position*df.Salience).sum() / \
(df.Clout*df.Salience).sum()
import pandas as pd
df = pd.read_csv('greece.csv',sep=',')
df = df.sort_index(by='Position')
print df, '\n', 'mean voter', mean(df)
print 'weighted mode', weighted_mode(df)
Stakeholder Clout Position Salience
0 Greece 100 40 100
10 USA 30 60 100
1 IMF 20 70 100
2 Juncker 10 80 100
3 Germany 100 100 80
4 France 70 100 100
5 Dijsselbloem 70 100 40
6 Tusk 70 100 40
7 Italy 50 100 100
8 ECB 20 100 70
9 Spain 20 100 100
mean voter 82
weighted mode 100.0
Shocks
Latest version also introduced a concept of "shocks" by sometimes randomly altering some salience values, or dropping actors out of the game to see if the optimum point is "stable", if these changes effect the final outcome. BdM says he had to introduce this trick after a project he did analyzing Clinton's Healthcare reform bill and failed to foresee one scenario, he has since fixed the problem. The details can be found in Predictioneer's Game.
Inborn Ability
Some ppl do have natural skillz on predictive social dynamics meaning basically an ability to run BdM's computations in their head. When BdM put the algo up for competition against international politics analysts, he saw some people consistenly tied the machine. Such people, then, basically had that internal machinery for social dynamics prediction wired into them.
Links