View Full Version : evolutionary dynamics in the US Bond

03-18-2002, 02:16 PM
(You will need your own daily bar-chart of the US 30-yr future going back a couple years to enjoy this post.)

As you might recall, last November the Bond put in a huge upside spike and downside reversal that played havoc with the purely-systematic trend traders. Dunn, the most straight-up/ABC trend follower of them all, with a billion dollars under management, got hammered to the tune of down over 23%. Other players, like Hawksbill, escaped the executioner only through the fortunate application of discretion overlaying their systems. Meanwhile, Bill Eckhardt - with near a billion under management but governed by his "chop filter" - basically sat on the sidelines.

One common thread running through all my posts has been that the environment dictates the kinds of traders which evolve. As such, when the environment changes, and you know what kind of traders are living in it based on previous environments, you can very crudely predict the texture of the chart in the near future, based on the errors of existing traders flailing in a vacuum of the anomaly they have evolved to exploit, and new anomalies sitting there under-exploited. It would seem such a major event in the bond would have to echo in the form of such a detectable change in the texture of the chart.

A quick look at the chart will confirm this hypothesis. From January 2000 - the epicenter of serious losing period for a lot of system traders - through November 2001, you had a fairly uniform chart texture in the bond. If you consider A to be the 1-day price range, B to be the 5-day price range, and C to be the 15-day price range, over this period you will notice that A and C were large relative to B.

In other words, traders who entered or exited too quickly or too slowly woudl have gotten chopped up, while the traders who used, say, a 5-day move to extrapolate a 15-day move should have grown steadily over this period. And that is just what we saw, with the real simple, mainstream traders like Dunn and Hawksbill being up over 100% over this period. But if you take a trader who uses a large A as a contrary indicator on B - a chop filter - you would have someone who basically sat on the sidelines and didn't grow at all.

Now consider the events of November. To a trader who uses a large A to stay on the sidelines, the big upswing and reversal woudl have eben invisible. To a fast-entry/fast-exit trader - who uses A or A+1 as positive-coefficient indicator - the move would have been viewed as two profitable trends, an uptrend which they could have gotten in at the front end of, and a down trend which they would have gotten in at the front end of. For a medium timeframe trader, November was buyign at teh high and selling at the low, and for a very long or slow trader, it was still big enough to catch them at both ends, but less severely.

So what we should have had is an expansion of traders who enter based on - and therefore exacerbate - short-term moves, and a contraction of traders who exacerbate medium-term moves. Meaning, the A-based traders should begin to expand B relative to A, and the B-based traders should contract C relative to B, more or less. And that is just what we saw in Decemeber, Jnauray, and, most of all February - typical-to-large 1-day range, a huge or proportional 3-day range, and a relatively small 15-day range.

So this would seem to explain why Bill Eckhardt, after sitting on the sidelines basically for two years, finally came in and took his beating. Both the fast traders, and the traders who use A relative to B as a sit-it-out filter, came in like crazy and got their heads handed to them. Meanwhile, for the slightly bigger-picture traders, most of this action was invisible, and they only lost a little money.

So, now for the question, assuming this persists, what will be the next texture? In my opinion, that will depend on whether any large trends actuallyu emerge. If not, the medium and long-termers will stagnate, while the fasties and arbs contract, leaving the environment under trader for the next trend in any timeframe. If a large trend does emerge, the fasties will chop themselves up around it, until the fade and teh arbs becoem valid again, while the medium-termers will do so-so and expand at the long-term rate.

In short, see who does well and evolve proactively or in anticipation to fade chop in their timeframe and, whoever contracts, look to fulfill their role the next time their kind of move emerges.


03-18-2002, 10:16 PM
am not quite sure what evolutionary dynamics means...but when the govt says they ain't gonna issue any more long bonds - thats evolutionary. Now this was unexpected news and came as a shock on Oct 31. The market responded wildly as this news was digested in the investment community. A great example of efficient markets. Not a great example for charting elroy -

03-19-2002, 09:34 AM
First of all, you concede that there are correlated pockets of invisible buyers and sellers whose arrival is gradual across time. Were it not for "trend traders" attempting to extrapolate immediate price changes, charts would be serially correlated.

Next, it follows quite clearly that, if all these trend traders are evolving independently, and yet they are reacting to the same signals on the same charts which also reflect the price moves created by their actions, that they must evolve a complex handoff and cooperation process, which can be considered an ecosystem.

As such, it is easy to visualize how, with such a huge upswing and reversal as we saw in November of 2001, the traders who had evolved to extrapolate the first fast blips in either direction would have expanded in wealth, while the traders who then take a handoff of their inventories, if the first blip holds, would have contracted in wealth, by buying at the high and selling late on the way down.

If we assume these fast traders are people who enter when some move in less than, say, the 5-day timeframe gets large relative to the range in a shorter timeframe - and if we assume these slower traders are traders who enter when, say, the 15-day move gets large relative to the 5-day range - it then makes sense that the 5-day and shorter ranges would widen - as the overpopulation of fast traders chops itself up and down in a vacuum of moves like the one huge one they evolved around - relative to, say, the 15-day range.

When you say it was a great example of efficient markets without being more specific, I cannot imagine what you mean. What was the enormous change in information between November 1 and November 15? I would argue that, quite the opposite, the reaction was anything but efficient, and the reason why was the magnitude of the news, combined with a one or two unfortunate chart coincidences.

As I stated, for traders to evolve independently and yet cooperate, they have to each learn some precision of sorting as to what signals they will take and pass up, so that a sort of spontaneous division of labor forms. Considering intraday S&P's, some traders may buy or sell the first blip of a move, the source of which is generally kept as concealed as possible from Wall-St. friction-desk operators.

If the first up-blip holds, the inventory may be handed off to a group of traders who buy an absence of pullbacks. If it continues on upwards, the inventory may be handed off to "breakout" traders who buy moves that are large relative to recent trading ranges, or relative to the time of day. Some percentage of the population of extrapolators learns to step off at off-peak times of day, where the underlying correlated pockets of buyers and sellers are generally not as big.

Most news in the S&P's is scheduled so, quite often, the shortest or fastest-trigger extrapolators will attempt to trade the 10:00 AM ET numbers, but the 5-minute timeframe traders will step back, in favor of the manned desk traders, who have better direct visibility into contingent insitutional demand based on alternative news outcomes. But after 9/11 you had this enormous post-news trendiness, so very quickly everyone learned to jump on the novel stream of unscheduled news.

As the trendiness faded, but the anthrax news kept coming at random times, it created this problem where news buyers and chart extrapolators would both hop on at the same time, as they hadn't had time to evolve a division of labor, or sort between the appropriate collective size of participation. And so it would trend after the news hit, as all the extrapolators hopping triggered each other, but were smeared out across time, and then the market would reverse.

So what we had in the Bond last November was an existing uptrend, which meant all the longer-term trend followers were already long, when an outsized or uncommonly large news event hit. The expansion of the daily range...

I won't go into detail, but you get the idea. Efficient means a bunch of blind robots who, if they get strobed with some chance, non-standard combination of stimuli, will fall out of step with one another, so that some collide and die. Afterwards, cooperation along pre-existing lines between the remaining or new relative assortment robots will produce a different texture of chart.

If, by efficient, you mean that when the big Bond news hit on October 31, 2001, that everybody knew exactly how many bonds everbody else now wanted to buy, and they were able to coordinate and match off inventories in anticipation of one another non-redundantly, you couldn't be more wrong. But I know you can't mean that:)

So what did you mean when you called it "a great example of efficient markets?"