The default prediction that we’d make using economic theory (or common sense) in the absence of game theory is that, in the turbine generator case, General Electric should have undercut Westinghouse because the former has lower costs.
But we start to see why it didn't when we introduce capacity constraints into the Bertrand model. Capacity constraints can stem from two things: decreasing returns to scale, or demand-uncertainties that create expected inventory-cost risks and induce firms to produce to order and carry backlogs. The second condition certainly prevails in the turbine generator case. Anti-trust regulators observed that both GE and Westinghouse were pricing higher than conventional economic analysis predicts. So they assumed collusion and prosecuted, not just once but several times.
In fact, as Ghemawat's research shows, the firms tried to collude but couldn't manage! Why?
The firms were looking for a pure-strategy Bertrand Nash-equilibrium, which doesn't exist, leading instead to so-called ‘Edgeworth cycles’. This Bertrand game has only mixed-strategy NE (like the river-crossing game), just as Ghemawat’s data on the turbine generator case suggest.
Therefore, the firms could achieve tacit, or non-cooperative collusion. This sounds like a contradiction in terms, and that's how governments tend to see the matter. But governments are simply wrong about this. Non-cooperative collusion can occur whenever there are Nash equilibria at which prices are higher than they are in competitive equilibria. Firms can find these without necessarily having to gather in smoke-filled rooms, by using signaling and screening in their games with one another. The turbine generator industry is a case where that's exactly what we'd expect. (So GE and Westinghouse managers would be correct if they felt that they were subjected to unjustified legal harassment.) The models:
Let capacities be denoted by xi (i = 1,2)....