Daily Speculations The Web Site of Victor Niederhoffer and Laurel Kenner

Home

Write to us at: (address is not clickable)

James Sogi

3/31/05
Do Bears Sleep at Night? An Experiment

It has been claimed that volume has no predictive value. If so, there should be no difference between the night session and day sessions. So just for fun, making 705-minute bars, a day bar and two night bars, showed some interesting new patterns and led to the following experiment.

Dr. Brett talked about this once last year, but it seems that the night prowlers might have something different going on than the sunshine gang. Chair and the Senator often speak of holding overnight. Some counting shows that since 94 the mean overnight move is slightly positive averaging .65 pts with sd 7.7 and an apparently significant distribution looking at only the 789 non zero nights. Even more interesting, as Dr. Brett found last year, during the bear market 2000-2002 the average overnight move was 1.2 pts positive sd 7.22 out of the 310 non zero moves which was also significantly distributed. It looks like bears sleep at night, especially when they've had plenty to eat during the day.

Add the morning session to the last 1,000 trading nights and the average gain is .94 pts sd 9.94 with significant distribution for the 791 non zero occurrences. So it seems bears not only don't eat as much at night, but they sleep late.

One of the dangers of hunting at night is identifying the target properly. For months the black boars in the black night proved elusive with their great sense of smell, night vision, and speed. One edge over them was they could be heard rustling and snorting quite loudly from a good distance. But you still need to get close for target identification and safety issues. There is no room for mistake. They could see me and my flashlight coming a mile away. But when I got an infrared night vision scope I had the edge over them.

With this in mind I foraged through at time and sales data with an eye to identifying who is behind the Globex bids and offers. Is there identifiable behavior? The 'Flipper' has been discussed and he has been spotted by bid ask behavior and size. There are a lot of of big bids, but fewer actual trades, so games are being played. A lot of posing, rustling and snorting is going on. Is this quantifiable? Are there quantifiable anomalies going on on the time and sales order flow? What, if anything, would these reveal? Nasdaq II has its market makers labeled. Who is the masked man behind Globex?

Its good to know when you are the hunter and when you are the hunted. Is that 3000 bid a bull or is it a bear dressed up as a bull to lure in an unsuspecting calf to the slaughter -- the Judas steer? This is just a start but here are some ideas on things to count. Much work is left to do. This might help in the foraging market ecology issue. When do we see T-rex show up at the watering hole. What is his footprint? Do the computers, T-rex's, Flippers, locals, hedgies, 1 lotters, have certain repetitive styles. Grandmasters beat computers. The computers may not have a defensive function built in. Are the elephants predictable ala Labagola? The goal here is not to speed grind, but to see if cumulative totals have any predictive value over time for speculative purposes.

Things to count

Ratio or differences of:

bids hit/offers hit
best bids#/Ask#
bids withdrawn rate /ask withdrawn rate
# trades at bid / # bids
# trades at ask / to # ask ratio
trades at bid / trades at ask ratio
# of trades per hour / Volume of trades per hour
ratio of best bid to market depth
distance from best bid to market depth bulges
rate of change on bid size or offer size

Effect of these ratios on subsequent prices on various time scales. This is more than market profile.

Philip J. McDonnell comments:

Some years ago I listened to a computer science seminar given at the University of Washington. The topic was using genetic algorithms to design robots. Essentially the idea was to write a program which would design and test robots without human intervention. The success of any robot design was measured by how well suited it was to forage for rewards (food) in the small one room virtual environment in which it lived.

The genetic algorithm essentially meant that designs were generated at random. Arms, legs and tuned eyes as well as other attributes could be added or deleted at random in each generation. If a given robot failed to find rewards above a certain threshold (starvation) its design was not used as the starting point for a subsequent generation - its line died out. Testing was done by another program written for such purpose and allowed many thousands of generations of designs to be tested without the necessity of actually building the physical robots. Successful robots passed their basic design on to the next generation with slight random alterations. After many thousands of such generations and hundreds of restarts to see what variations might occur a few key principles emerged.

The researchers immediately noted that the successful robot designs all narrowed in on bisymmetric guidelines. Some designs were bizarre gothic looking winged structures which no human would have ever conceived because they served no purpose. Even these were bisymmetric always evolving to a balanced 2 or 4 wing design. Most designs were rather prosaic 2, 4, 6 and 8 legged creatures with eyes. The researchers had correctly anticipated that bisymmetry was a likely outcome of their experiment.

The researchers also noted something extraordinary which they had not anticipated. The designs began to bifurcate into two quite distinct lines. One group became light colored, generally fewer legs with eyes which were tuned to daylight. The other group evolved along nocturnally adapted lines with eyes optimized for low light or no eyes at all and often more legs for stability in low light conditions and darker colors. The researchers had not anticipated this development at all. It took the genius of a random number based genetic algorithm to explore the possibilities.

The parallels to different classes of market participants are clear. Some are adapted to overnight trading. Foreign orders often shows up at the open in the US markets. Depending on how cheap stocks are in euros, pounds and yen and other factors the foreigners will either be net buyers or sellers on the open. During the day market makers and hedgers operate but prefer to square their books by the close. So if 2 at the money calls are purchased from a market maker he will often immediately buy 100 shares of stock at market as a hedge. He may be offering a .10 spread on the calls and is relying on the fact that there is usually a .01 spread in the stock. He is evolved to trade during the day only when his stock is open. The markets can support both day and night traders quite well.

Henry Carstens comments:

Artificial Intelligence, A Modern Approach by Russell and Norvig has a section on "Adversarial Search" that relates to the above scenario and to trading in general. Each section of the book also breaks out the current state of learning algorithms for various problem types. Very good for generating not just ideas but entire research frameworks.

Also, Tufte's Envisioning Information  has multiple examples of displaying time-based information that have helped me reformulate all my paper-based trading aids. Tufte's books have been a valuable resource of ideas for me lately.

 

James Sogi is a philosopher, Juris Doctor, surfer, trader, investor, musician, black belt, sailor, semi-centenarian. He lives on the mountain in Kona, Hawaii, with his family