©2006 All content on site protected by copyright
Victor Niederhoffer on Negativity and the Markets
A few of the women on the speclist, especially Debra Moon, have suggested that by analyzing the contents of this list, they could understand the market. There are many fruitful extension of this rich idea and they inspire the following.
One must start with the well known negative serial correlation of short term moves in stocks, the canes et al. Also the tendency of most comments to follow the price, with negativity increasing in relation to the past stock move down. Everything has to be adjusted for these, but leaving them aside for the moment ...
One hypothesizes that the more negative the content of posts on a list, the greater the positive expectation going forward. This would be measured by the number of synonyms for good and bad in the posts, according to scales and calibrations contained in the work of Osgood. The same could be done for a column almost invariably as negative in tone as the bifurcated -- let us say enigma -- that came from us. Its base level is perhaps 95% negative as measured on the Osgood scale. But what about when it moves to 99%. or 91%? The former would occur in those one in three months say that the market goes down, and the latter would occur in a time when the market is at say a 9 month high, like above 1340 in the S&P when all shorts are suffering from squeezitis. So one's predictive regression would also have to take account of the past market move relative to its 90 day high or so.
Putting that aside also, one would like to test the negative content of news and reactions adjusted for past price move, the natural tendency to complacency, and the varying degrees of proneness to the overly favorable self reported evaluation of greatness biases, all as a predictor of future market moves. This could be classified by source, i.e. newspaper, email list et al. This is a nice problem in the un-tangling of hypotheses for which the confidence profile method, a method based on the multiplication of likelihoods, would be useful.
Big Al adds:
One thinks that Yahoo Finance headlines might be such an indicator. Anecdotal and imaginary, but it often seems that the day goes something like this:
By 10:30, S&P moves +5 pts
Yahoo Finance headline appears at 10:30: "Stocks gain as oil prices drop"
time to go short
By noon, S&P moves back to par
new Yahoo Finance headline appears: "Markets give up early gains on XYZ earnings"
time to go long again
At 2:00, S&P back to +6
Yahoo Finance headline: "Traders buying on positive inflation data"
time to close out
It just seems that it takes about the length of some moves for the news writers to "see" the move, write it up and get it posted.
Jim Sogi mentions:
Just as the number of trades carries more statistical significance than the sheer volume, the number of posts on various lists might carry more information than analysis of content. The content may be subjective but it has appeared that the main list, considered to be mainly bullish, is quite busy when markets are up, but at dark bottoms become almost silent and has been a good indicator for market operations. Another good experiment would be to find a list with a bearish tilt and count the number of posts at recent multi month highs and consecutive multi day highs. This is a methodology to test a contrarian indicator since the market is most bearish after numerous multi day highs, and new monthly highs, and most bullish at dark monthly lows. Underlying this method is the natural herding instinct and the reason markets tend to trend up, then down in cycles.
The May down cycle was 6 weeks and 8%, and the recent rally was about 8 weeks and 8%, and whether random or not, retrospectively creates the appearance of a cycle. The perfection and beauty of the wave over so many weeks and months as opposed to the apparent cyclic formations of a random walk , by eye at least, indicate to this observer that more is at work than random forces. If there are larger forces at work it would give a great advantage to a speculator to know by simple time measurement the time for a turn. In any case it is better to buy within a week or two of the bottom rather than buying at the top and selling at the bottom. The measurement of time might have information as distinct from the measurement of price. After 8 weeks up and 8% up do the probabilities favor another 8% move up for the next 8 weeks? Sampling methods on weekly returns might provide an antidote to the insufficient data points. Using the Professor's Fourier analysis (like his work on lunar cycles) on the random samplings versus the actual might indicate whether there are larger cyclic forces at work that might be harnessed. Does the actual have a greater degree of cycles than the random? What is the length of such cycles?
A few more ideas: Watching market depth on CME it seems that very high depth, which indicates high liquidity, suppresses price movement and variability, and that a lower amount of depth leads to greater variability. A curious thing happened the other day when Globex went down for a short period, the market had a bit of cheesecake and showed 10 levels of depth rather than the normal 5 for a while, and gave a glimpse into the inner workings normally hidden, like a quick peek behind the scenes.
Debra Belanger Kettle comments:
sheer volume and levels of hostility vs. camaraderie/politeness, even snoozeville (as in boring) seemed noteworthy when I first mentioned my observation ... I was new to the list at the time and found the peace vs. conflict fluctuations of the posts quite fascinating.
I am not sure what it means though, or how to test ... whether ipso facto or post facto it is something significant.
Dr. Brett Steenbarger offers:
I do think this would be a very interesting undertaking. My leaning would be to first examine grosser relationships, such as the frequency of posting vis a vis recent (and prospective) price change. I would also be tempted to examine the relative frequency of different kinds of posts (analytical ones, personal ones, etc) in that vein. Yet another measure would be the number of different threads and the extent of participation by various list members.
My preliminary hypothesis would be that people are more likely to post to the list and participate in threads during periods of heightened uncertainty. Posting, in that vein, would be seen by a psychoanalyst as a higher order mode of coping: a way of trying to make sense of ambiguity. One of the famous measures of coping styles breaks down the ways people deal with stress into three broad categories: problem-focused coping, emotion-focused coping, and avoidant coping. It would be interesting to categorize posts similarly, viewing the List as a social medium for dealing with uncertainty.
Freud postulated that, under conditions of duress, we regress to lower (developmentally earlier) forms of coping. Normal problem-based coping might regress to emotional or avoidant modes. One might expect market inefficiencies to be greatest during times of such regression.