Tail event: Difference between revisions

Notice how arbitrary that “relationship” between two discrete transactions is. If they are independent of each other — and in a first order sense, they are — the participants in the later trade do not know who or where the participants in the earlier even are, let alone what their motivations for trading were — then a “trend” we draw between them is more or less meaningless.  

But we draw it because we make some assumptions about the homogeneity of all market participants: we assume all have more or less perfect price information, and that all are propelled by an essential economic rationalism: you don’t sell things you expect to perform better than comparable investments, and you don’t buy things you expect to perform worse. Each investor’s private motivation is nuanced and personal — how is the rest of it's portfolio placed, what are the local macro risks to which it is especially sensitive, but largely these proclivities cancel themselves out in a large sample and we can treat professional market participants as largely homogenous.

But there is a second order sense in which the earlier and later trades are related: the later participants know about the earlier trade and its price — and its position in a trend from the trade before that — and they will use l this abstract information to form their bid and ask. This interconnectedness of all similar transactions means they are ''not'' independent, as the probabilities of [[normal distributions]] require, but most of the time it's close enough: the immediate transaction history is pretty chaotic — as traders say, “noisy”— in the immediate term, here the dissimilarities between trader motivations are most pronounced, but over a large aggregation of trades these dissimilarities tend to cancel themselves out. A “signal” only emerges from the noise over time. If all traders are using market information, this immediate interdependence looks like independence. So a “normal” probabilistic model<ref>I am working hard not to use the intimidating term [[stochastic]]” by the way.</ref> works fairly well. In the same way, when you bounce a ball friction, energy loss, structural imperfections and intervening causal interference means the conditions to fully satisfy Newton’s model are never present, so a bouncing ball never quite obeys it, but it is close enough. So it is supposed to be with statistical techniques for measuring behaviour in the market. The occasional intervention of idiosyncratic behaviour is basically noise.

This is not just the crowded theatre phenomenon, when everyone stampedes for the exits at once, and the narrow aperture makes the stampede all the more urgent, and therefore dramatic — but second order features. An investor long “on margin” might wish to, and be able to, ride out a short term crash by meeting margin calls. In most dislocations this is the obvious and — if you can manage it, correct — thing to do. The market usually recovers, at least in the short term. But meeting your margin call means drawing on your revolving credit facility and your bank is experiencing a liquidity crisis and unexpectedly pulls you lines, or suspends withdrawals, as a result of its own market exposure to the crash. Your prime broker, usually patient with you and tolerant of peripheral looseness in your margin operations, is also under pressure, has told you today there is no flex, and for good measure, it is jacking up your IM.

Cash is suddenly King Queen, Jack and Ace. People on the TV wandering dazedly around outside their buildings clutching [[Iron Mountain]] boxes full of personal effects.

buy or sell. In the modern days of computerised trading everything is very clean, tidy observable, unitary and discrete.