Their model has a set of agents with a 'talent' normally distributed between 0 and 1 (mean 0.6, SD 0.1).
Each agent gets a random series of doubling and halving events. The doubling is applied with probability = 'talent'.
The lengths of the series of events have an exponentially decreasing distribution (four longer is x10 less probable, approximately).
The application of these series of events to the agent is supposed to simulate a life history of random opportunities, exploited according to talent.
BUT...
Half the events are losses, half the events are possible (P<1) gains. So for an agent with average luck (which is every agent in the long term), their capital decreases with time. The simulation has been rigged so agents can only win with luck.
It's actually worse than that. If we take a population with only median and maximum talents, the probabilities for each event are: halving: p=0.5, doubling
because median-or-above talent: p=0.5*0.6. Meaning that in 80% of the events for a given actor the talent is not considered at all.
So the result that luck plays a larger part isn't an emergent property of the model, it was explicitly baked into the model from the start.
It wouldn't be unfair to call this paper fraudulent.
Their model has a set of agents with a 'talent' normally distributed between 0 and 1 (mean 0.6, SD 0.1). Each agent gets a random series of doubling and halving events. The doubling is applied with probability = 'talent'.
The lengths of the series of events have an exponentially decreasing distribution (four longer is x10 less probable, approximately). The application of these series of events to the agent is supposed to simulate a life history of random opportunities, exploited according to talent.
BUT...
Half the events are losses, half the events are possible (P<1) gains. So for an agent with average luck (which is every agent in the long term), their capital decreases with time. The simulation has been rigged so agents can only win with luck.
DOI: 10.1142/S0219525918500145