Goldman Sachs Black Box Problem
By Lipa Roitman Ph.D.
December 20 2012
Nobody knows what’s going on inside the Goldman Sachs GS. It’s a black box problem for most of us. One recent article attempts to analyze GS P/L statements and is discussing the cost cutting efforts to reduce headcount and invest in technology; it says the stock is overvalued, similar to another earlier article. However still another article is looking at the same numbers and describing GS in a positive light.
We know that GS is using mathematical models in their trading strategy. Now we put its stock under the same analysis. Can GS stock be modeled?
In an effort to quantify the GS value Ivan Kitov recently described a model of GS price based on the concept of the link between consumer and stock prices. The new article is an extension of his previous research in which he found that Exxon Mobil XOM and ConocoPhillips COP stock price could be predicted using the difference between core and headline CPI in the United States. He found that linear trends in the CPI difference allow accurate prediction of the prices at a five to ten-year horizon. His new extended model of GS price connects the share price as a weighted sum of two individual consumer price indices selected from a large set of CPIs.
Here we describe another model for stock valuation that is different from Kitov’s. It is based on the realization that a stock value is a function of many factors which interact in a non-linear way and affect the future trajectory of the stock creating waves in prices. Being completely empirical, the I Know First self learning algorithms analyze the inputs and rank them according to their significance in predicting the target stock price. Then they create multiple models, and test them automatically on the historical data. The robustness of the model is measured by how it performs in different market circumstances. The best predicting models are kept and the rest are rejected. Such refinement has continued daily as the new market data is added to the historical pool.
Fig. 1 shows the I Know First model results in predicting the fall and rise of GS stock from April to November 2012. The basics of this stock forecasting software are described here, here and here. Each point on these charts was taken from the actual daily forecast published the morning before the next market open. Each forecast contains six different time horizon forecasts, from three days ahead to one year ahead. The charts show the actual price in blue and the signal line in red. The positive or negative (up or down) signals of the forecast were added to the actual last known price at the time of forecast. Thus, when the signal line is above the actual line, it means “buy”, if below, then “sell”. The green and red arrows show what would be the best times to enter the market.
.
Conclusion
The value of GS Stock as a function of time can be adequately predicted using a multivariate model based on historical stock performance relative to the rest of the market
