Search for probabilities

January 15, 2016 09:00 PM

Thinking in terms of probability is a central tenet of being a Superforecaster. Yet, when when it comes to projections, all too often forecasts are delivered in 30-second sound bites or a tweet. We reached out to EidoSearch, a forecasting firm that uses a patented pattern matching technology, to project event outcomes and find relationships in Big Data.

When examining data, EidoSearch President and Co-founder David Kedmey says investors should take a disciplined approach to discover a better understanding of future events. 

“There are three questions: First, what does it mean to make an investment decision? Next, what’s the most likely outcome for an investment and what are the range of outcomes?” Kedmey says. “Finally, what is the probability of successful investment? That third question is the hardest to answer.”

EidoSearch aims to answer those questions and provides a deeper assessment of the ranges and probabilities of forecasts issued by Wall Street’s most prominent analysts in both 2015 and in 2016 (see “Range of outcomes,” page 23). Using a conditional probability assessment, investors can better understand the likelihood of a projection issued by a 30-year Wall Street veteran or your uncle who thinks that Apple stock is heading to $200 by the end of next year.

“The big idea is how to use a world of data,” says EidoSearch CEO and Co-Founder, Dr. Steven Zhang. “Our methodology is data-incident based modeling (DIB). Think about how people look at things. 

For example, look at Apple stock. If you ask a series of questions—why people think a certain way—they will base it on past experience. They know if they have seen something before. From experience, they look at criteria and say that Apple should go here.”

The process relies on collecting as much data as possible. First, EidoSearch identifies a data series in any historical data set and searches for multiple patterns. Next its search technology identifies similar conditions across the data stream and returns the most relevant results. Then, the technology analyzes the results and generates a probabilistic range of outcomes that could not be produced by other modeling techniques. 

“People try to make predictions in terms of probabilities,” says Zhang. “Everything has a probability, including the weather forecast. From experience, a person has seen a scenario before. Now, technology allows us to use a computer to compute these conditions. We can look through history [and find] all of its conditions, all of its patterns; and situations are matched.” 

Naturally, that concept of history is to understand rare events that could lead to a significant downturn or surges in the price of a particular equity or commodity. 

“There are two important aspects to what we do when looking at events,” says Zhang. “We can factor in the probability of a rare event. This hasn’t happened many times, but it has happened before. You have to have seen this type of scenario before to make that type of prediction. We then quantify that prediction in order to give a statistical representation of those events. Next, using our methodology, we look at the stability of the historical distribution. We can say, ‘here is the prediction,’ but know there is still uncertainty based on the distribution.”

Although technology and academic research are advancing the importance of thinking about probabilities, Zhang acknowledges that barriers still exist. 

“Human nature is to want something certain. We want better organization, but it’s kind of lazy to simply want to know the outcome. But probability is based on knowledge. You can say that there is an 80% chance of rain, but people just want you to tell them whether they should bring an umbrella [or not].”

 

“Range of outcomes,” below, provides an assessment of equity and commodity markets based on a conditional probability assessment for 2015 and 2016. All 2015 projections are based on data available on Jan. 1, 2015 and provide an assessment of the probabilities of certain price targets on Dec. 31, 2015. 

For data projections in 2016, all data was analyzed during the first week of December and offers the probabilities of certain price points during the first week of December 2016. The following is a range of probabilities on where the U.S. unemployment rate will sit during the first week of 2016. When looking at the top figure—the percentages listed—each signifies the probability that the associated figure will be at this level of unemployment or less. 

This means that there is a 40% chance that the unemployment rate will be at 4.52% or less during the first week of December 2016 It also means there is a 60% chance that unemployment will be at 4.52% or higher for the same period.

EidoSearch also breaks down each figure in terms of 10% probability buckets. Based on the figures below, there is a 10% chance that unemployment will fall between 4.6% and 4.7% in December 2016, just as there is a 10% chance that it will fall between 5.04% and 5.74% and a 10% chance that it will be above 5.74%.

Range of outcomes

The Federal Reserve’s decision to not raise interest rates after the end of its third round of quantiative easing was one of the most important stories of the year for the markets. In the second week of December 2015, the 10-year sat at 2.12%, well below the expectations of every major analyst projecting in Barron’s or in other major media outlets. Interestingly, EidoSearch’s conditional probability expected that there was a 50% chance that interest rates would be at or below 2.32%.

Meanwhile, every projection made in Barron’s in 2015 was in a less probable range for the year. Subramanian and Lakos-Bujas were the only two analysts whose projections fell in the 70% to 80% band, meaning that the probability of their rate estimate was a little less than 25%. Meanwhile, Praveen at Prudential fell in the 90% range of estimates. There was only a 10% probability that rates would be 3.25% or higher at year end, meaning his estimate of 3.5% faced even greater odds.

Each December, media outlets corral the top analysts on Wall Street and ask for their projections on how the S&P 500 will perform by the end of the following year. Barron’s releases its annual roundtable, while other sites like Business Insider accumulate projections for its readers. The S&P 500 was just below the 2,100 target set by three prominent analysts with three weeks to spare in December. But other analysts significantly overshot the year-end figure. As this spectrum by EidoSearch explains, there was a 50% chance that the S&P 500 would sit at 2,218 based on a probability assessment of data compiled on Jan. 1, 2015. 

This spectrum also points out that there was a 30% chance that the S&P 500 would be at 2,093 or lower at the end of the year (in the range where the figure sat at the time of publication). 

Based on its 2015 outlooks released at the end of last year, Stephen Auth of Federated Investors was the most bullish analyst with a target of 2,350. That figure was just below the 70% band listed above. There was roughly a 31% probability that the S&P 500 would hit or surpass Auth’s 2015 target.

Looking toward 2016, EidoSearch projects there is a 50% chance that the S&P 500 will be at 2,365 or higher for the year. Meanwhile, there is a 40% chance that the S&P 500 will be at or below 2,300 during the same time frame. That figure represents the highest projection from the nine analysts whose forecasts we examined this month.

Peter Schiff at Euro Pacific Capital has said that gold prices are headed to $5,000 over the long term. But statistically, the odds of it happening in 2016 would signal levels of unimaginable inflation and no resulting societal collapse that would place the importance of bread over the importance of non-edible gold.

Based on historical data, EidoSearch projected that there was a 40% chance that Apple stock would be at or above $112.32. The stock price of roughly $115 per share during the first week of December is within the 40% to 50% range projected by EidoSearch’s models. 

Based on historical data, EidoSearch projects that there is a 50% chance that Apple stock will hit or surpass $125.72 in 2016. The bullish call that Apple stock is poised to hit $200 within 12 months faces steep odds. The probability of Apple stock hitting $192.12 or greater sits at just 10%, according to a conditional probability assessment.

The decline of oil prices to near seven-year lows was unlikely, according to Eidosearch. The WTI price of nearly $37.20 was approaching the range of 10% or less in the final three weeks. Eidosearch’s probability assessment had set a 50% likelihood that oil prices would be at $54.73 or lower for the year. But below $38 fell near the 12% range.

While many investment banks, government agencies and research firms assess the price of annual price of oil in terms of averages, EidoSearch’s 12-month outlook sets an expectation for a specific day and price in the future. The 14 contributors to this year’s Modern Trader forecasting issue all set their expectations of where crude prices will fall during the first week of December 2016. As noted, Rod David is very bullish on crude prices, yet a conditional probability assessment shows that his price expectation faces steep odds. In fact, all 14 analysts project that oil will be above $40.16, which EidoSearch projects will be the center of the distribution. This assessment will be intriguing going forward given that OPEC could dramatically change its production policy and mountains of debt could push many shale companies out of business, suppressing the glut of oil in the United States.

About the Author

Garrett Baldwin is the Managing Editor of the Alpha Pages and the Features Editor of Modern Trader. An author and Baltimore native, he earned a BS in journalism from the Medill School at Northwestern University, an MA in Economic Policy (Security Studies) from The Johns Hopkins University, an MS in Agricultural Economics from Purdue University.