The article by Gillian Tett and Peter Thal Larsen called “Market faith goes out the window as the ‘model monkey’ loses track of reality” talks about how people that work in the banks and who trades on the market only after entering inputs into the model were called F9 model monkeys, companies and major banks using complex mathematical modeling systems to trade on the market. It emphasizes on the problem that some hedge funds and banks experience such as trading losses due to the fact that these institutions are basing their business strategies on the bunch of complex models that are not accurate. Companies like Ford and General Motors are faced with the credit downgrades because they rely on the banks that heavily use mathematical models that were either created by the theory of the Nobel Prize winning economists or are partly untested and not on the reality and the behavior of the real life. The article also talks about how banks started developing their own models like “senior” and “mezzanine” models for figuring out the collateralized debt obligations (CDOs) pricing and to selling them to the clients. Banks sell simpler models to the public and keeping the models that are more risky and complex for them. Different banks have different models for the CDO pricing and market prediction, thus causing financial institutions like JPMorgan and Deutsche Bank a financial damage.
Discussion Question: Why do you need to be careful about what your model is telling you?
The reason why we need to be careful about what the model is telling us is because in the article the models that were used for market prediction and CDO pricing were all based on the regression analysis which was based on the linear regression. What regression analysis does is it enables us to develop a model to predict a value of a numerical variable in this case it is CDO pricing. The variable that we wish to predict is dependable variable that is CDO pricing, and variable that we use to make our prediction is independent variable that is all the debt and equity products, complex mathematical models and formulas and the hard work of the economists. So, the fact is that there are always unexpected changes on the market and by relying heavily on the models alone it enables us to adapt to those changes, thus causing losses. We have to take into consideration the fact that there is always a room for errors, market changes, and demand changes, therefore values cannot be the same neither can the values in a regression analyses be the same, it cannot be simply based on the straight line the results will always vary around it. We can use a standard error of the estimate to measure the standard deviations around that line and the assumption of regression techniques to calculate the errors, but I guess it would be next step that the “F9 model monkeys will take.
The second article talks about stock prediction, how stock prices can be predicted based on the certain events such as if NFL wins super bowl the prices will go up and how stock brokers and investors are allured into the data-mining. This article written by Jason Zweig talks about how brokers on the wall street trying to find an explanation or a cause that predicts the returns on the stock market not realizing that because of the massive amounts of information that stock market generates the relationship that is found is purely coincidental which gets them into data-mining. The data mined numbers gets so irresistible and alluring that people invest billions of dollars to find hypothetical results that no one even knows if they will work in the real world. To prove the point of the data-mining Mr. Leinweber decided to do an experiment that was used to predict US stock prices based on the annual butter production in Bangladesh in the past 13 years. He was able to predict a US stock returns with 99% accuracy that by tossing in US cheese production and the total population of sheep in both US and the Bangladesh for the past 13 years. However, there is no explanation or a legitimate reason why US stock returns would be determined by the Bangladesh livestock returns, therefore the results are coincidental. There is couple of rules to prevent data-mining. First and most important is the results have to make sense. Second rule is to look at the data in separate pieces. And the last rule is to give it some time; hypothetical result will not last once they face the reality in the real world.
Discussion Question: What is the difference between correlation and causation?
Correlation used to describe the degree of relationship between two variables and causation describes the degree to which one variable causes the other variable. For example, in this article we can clearly see that a correlation does not automatically calls for causation. Certain events that happened that might be correlated like US stock prices and Bangladesh butter production, but there is no explanation that one causes the other. Although, it looks very appealing and we want to make it so the US stock can be predicted based on the Bangladesh livestock, and stock prices do go up when NFL wins the Super Bowl, but we can’t not until there is some kind of a scientific or statistical explanation that make sense to this relationship. People want to believe what they see, often believing what they want to believe regardless of the facts. They tend to believe that everything is like a chain reaction. Especially once they see a relationship that works they attach a cause to it without basing it on a logical reasoning. We just have to make sure not to get caught by the data-mining; we just have to be able to distinguish between what is real and what is just a mirage.
No comments:
Post a Comment