Printer Friendly Version      Email this Article

The Broken Clock & Other Timing Phenomena

There is an old saying that "even a broken clock is right twice a day." One purpose of this saying is to warn us about jumping on the bandwagon of a prognosticator without extensive research and thought. Those who would predict future events vary from the Con man, whose purpose is to gain illegally by a scheme that he knows to be false, to those well intentioned and well trained in elaborate research techniques. This training in research techniques does not preclude conclusions that are as erroneous as those of the Con man. This is particularly true when attempting to predict events that are based upon human behavior. The intention is different but the results can be the same.

Two major "con" examples come to mind:

1. A brokerage "boiler room" starts with 10 stocks that they think might become hot. They select several thousand names to call and offer each person 1 of the 10 stocks as a "red hot prospect." If 3 of the 10 stocks actually take off they recall those customers who bought them and offer them 1 of 10 new "hot" stocks. This process goes on until they have a small number of customers who have had several extremely good experiences by chance (unknown of course to the customer). At each call they get the customer to buy larger and larger blocks of stock, which generate larger and larger commissions.
   
   
2. A stock newsletter obtains a significant position in a small cap stock. Then they produce a special issue or two recommending that stock. The float of the stock is small enough so that buying by the readership of the newsletter can drive up the price of the stock. The newsletter then disposes of its shares at a large profit. It then recommends selling the stock because of some factor that they have identified. Since some of the readers sold as the stock increased or shorted on the sell recommendation, the newsletter is able to keep a proportion of its readership for quite a while.
   
   

We all know people who continually predict either glorious times or doom in a market. Based upon the cyclical nature of events, even allowing for extreme manipulation in the opposite direction, these people will eventually be correct. However, the time interval between the original prediction and the actual positive results can be so large that the opportunity costs in following that advice will be disastrous. The phenomenon that lets us selectively remember those events that fit our expectations, makes us consciously aware of only the most recent successful prediction, and not the long history of failures. To protect against this phenomenon you must record the date and prediction of each and every utterance in order to get a complete picture of the prognosticator's success rate.

Sometimes, whether by insight or accident, someone will come upon a variable or set of variables whose change can predict future market behavior. It is not necessary that those variables actually cause the subsequent behavior, only that they consistently precede it. This discovery can be effective for considerable periods of time only to subsequently fail in its use. There have been newsletter writers who have apparently zeroed in on what led to either market advances or declines only to fail miserably when conditions appear that reverse the market valuations. They can only see the conditions that led to their success and not those that portend change from it.

It is extremely important to recognize the difference between correlations identified and causes identified. This is found basically in the difference between experimental research, and correlational analysis that is the fundamental tool of economic research. The closer we are to experimental techniques, the more confidence we can have that causes have been identified. This is because in experimental research the researcher possesses two "powers" that the correlational researcher does not.

1. He can physically introduce and manipulate the independent (presumed cause) variable.
   
   
2. He can physically partial out or hold constant, i.e., control potentially confounding variables. Confounding variables are those that can produce the same outcomes as the independent variable and so confuse interpretation of the research results.
   
   

An economist might have a theory that introducing certain incentives into a market will produce higher output. It is likely that there are many such variables that will produce the desired outcome. The experimental model would require at least two markets that are identical on all variables except the one that the researcher proposes to manipulate and observe. What can he do? He cannot identify two identical countries and introduce the incentives in one and withhold them in another. He cannot identify and partial out all of the potentially confounding variables. He can either try to statistically manipulate and analyze data that already exists, or collect data as they occur and then statistically manipulate and analyze it. He cannot physically introduce and control variables as a laboratory researcher can.

The computer, which has given correlational researchers tremendous computational power has, in many cases, led to logically failed conclusions. There is a tendency to simply "stuff" hundreds of variable for thousands of cases into the computer and look for relationships coming out the other end. The problem is that the more variables that you put in the mix, the more "relationships" will appear by chance. This is why investigations should ordinarily be guided by specific predictions gleaned from a theory. These specific predictions are then "tested" by the analysis. This reduces to a certain degree the appearance of "relationships" by chance.

Remember all that is necessary for a correlation to appear is that variation in one variable "goes with" or parallels variation in a second variable. THERE DOES NOT HAVE TO BE A CAUSAL RELATIONSHIP for this to happen. For example if you gathered data on the number of pigs on Iowa farms and the number of students in graduate school from about 1950 to about 1970 you would find a moderate correlation between the two! Does this mean that pigs on Iowa farms produce graduate school students? Or maybe graduate students produce pigs in Iowa. It is simply likely that both are the by-product of a growing population! If you throw enough variables into the computer you will find some strange "relationships". Correlations that might make no sense to you but some person will use them to "explain" his economic predictions. These relationships might actually parallel predicted market moves in a particular direction. You might actually, but accidentally, make money by following the recommendations. But, as history shows, when real conditions change, the power of that newsletter's prediction will evaporate like so much smoke in the night.

Read. Study. Do not bet the whole bankroll.

Caveat Emptor.               

March 14, 2001

Harry J. Clawar Ph.D.
HJC@angelfire.com

Also by Dr. Clawar