Wednesday, February 22, 2012

Annealing the Information Market




When analyzing complex systems, applied mathematicians often turn to Monte Carlo simulations. The concept is straightforward. Change the state of the system by making a random move. If the new state is an improvement, make a new random move in a direction suggested by extrapolation. Otherwise, make a random move in a different direction. Repeat until a certain variable is optimized.

A commodity market is a real-life concurrent Monte Carlo system. Market participants make sequences of moves. Each new move is random, though it incorporates experience gained from previous moves. The resulting system is a remarkably effective mechanism to produce commodities at the lowest possible cost while adjusting to changing market conditions. Adam Smith called it the invisible hand of the free market.

In severely disrupted markets, the invisible hand may take an unacceptably long time, because Monte Carlo systems may remain stuck in local minima. We may understand this point by visualizing a mountain range with many peaks and valleys. An observer inside one particular valley thinks the lowest point is somewhere on that valley’s floor. He is unaware of other valleys at lower altitudes. To see these, he must climb to the rim of the valley, far away from the observed local minimum. This takes a very long time with small random steps that are biased in favor of going towards the observed local minimum.

For this reason, Monte Carlo simulations use strategies that incorporate large random moves. One such strategy, Simulated Annealing, is inspired by a metallurgical technique that improves the crystallographic structure of metals. During the annealing process, the metal is heated and cooled in a controlled fashion. The heat provides energy to change large-scale crystal structures in the metal. As the metal cools, restructuring occurs only at gradually smaller scales. In Simulated Annealing, the simulation is run “hot” when large random moves are used to optimize the system at coarse granularity. When sufficiently near a global minimum, the system is “cooled“, and smaller moves are used for precision at fine granularity. Note that, from a Monte Carlo perspective, large moves are just as random as small moves. Each individual move may succeed or fail. What matters is the strategy that guides the sequence of moves.

When major market disruptions occur, resistance to change breaks down and large moves become possible. (The market to runs “hot” in the Simulated Annealing sense.) Sometimes, government leaders or tycoons of industry initiate large moves, because they believe, right or wrong, that they can take the market to a new global minimum. Politicians enact new laws, or they orchestrate bailouts. Tycoons make large bets that are risky by conventional measures. Sometimes, unforeseen circumstances force markets into making large moves.

The music industry experienced such an event in late 1999, when Napster, the illegal music-sharing site, suddenly became popular. Eventually, this disruption enabled then-revolutionary business models like iTunes, which could compete with illegal downloading. This stopped the hemorrhaging, though not without leaving a disastrous trail. Traditional music retailers, distributors, and other middlemen were forced out. Revenue streams never recovered. With the Stop Online Piracy Act (SOPA), the music industry, joined by the entertainment industry, was trying to undo some of the damage. If enacted, it would have caused significant collateral damage, but it would have done nothing to reduce piracy. This is covered widely in the blogosphere. For example, consider blog posts by Eric Hellman [1] [2] and David Post [3].

While SOPA is dead, other attempts at antipiracy legislation are in the works. Some may succeed legislatively and may be enacted. In the end, however, heavy-handed legislation will fail. The evolution towards ubiquitous information availability (pirated or not) is irreversible. Even the cruelest of dictators cannot contain the flow of information. Why would anyone think democracies could? Eventually, laws follow society’s major trends. They always do.

When Napster became popular, the music industry was unable to fight back, because its existing distribution channels had become technologically obsolete. Napster was the large random move that made visible a new valley at lower altitude. Without Napster, some other event, circumstance, or product would eventually have come along, caused havoc, and be blamed. Antipiracy legislation might have delayed the music industry’s problems in 1999, but it will not solve the entertainment industry’s problems in 2012.

In the new market, piracy may no longer be the problem it once was. Consumers are willing to pay for convenience, quality of service, and security (absence of malware). Piracy may still depress revenues, but there are at least three other reasons for declining revenues. (1) Revenues no longer support many middlemen, and this is reflected in lower music prices through free-market competition. (2) Some consumers are interested in discovering new artists themselves, not in listening to artists discovered on their behalf by record labels. (3) The recession has reduced discretionary income.

It is difficult to assess the relative importance of disintermediation, behavior change, recession, and piracy. But the effect of piracy on legal downloads is probably much less than thought. This may be good news for the music industry. After many large and disruptive moves, the music market may be near a new global minimum. Here, it can rebuild and find new profit-making ventures. These are the kind of conventional “small” moves for a normal, non-disrupted market.

Other information markets are not that lucky.