A better way to do economic forecasting?

A trader works on the floor of the New York Stock Exchange.
Getty Images
A trader works on the floor of the New York Stock Exchange.

Is there a better way to do economic forecasting?

There's been lots of talk about applying machine learning, big data and crowd sourcing to the stock market. A new firm wants to apply the same ideas to economic forecasting, and it is claiming its approach generates economic forecasts in real time with a higher accuracy rate than famous Wall Street forecasters.

The firm is Now-Cast Data Corp, and its founder, CEO Giselle Guzman, came by the NYSE recently to demonstrate the program.

You know how dismal economic forecasting can be. First, the very methods used to generate the official numbers involve huge margins of error.

Take new home sales. It's compiled by the Census Bureau from sample surveys. It takes a month to get out. For the July 2015 report, the Census Bureau said new home sales were 25.8 percent above the July 2014 level.

Now read the fine print: They are 90 percent confident that the actual number is 22.6 percent ABOVE or BELOW that number.

That is a HUGE margin of error. You could drive a truck through the estimate. Why are the estimates so...full of holes? Because the sample size is fairly small.

You could get a more accurate number by surveying more builders, but you would need more money, which the Census Bureau doesn't have.

This is true of all government statistics.

Wall Street, in an effort to improve on this dismal science, has an army of analysts and strategists who generate their own estimates of economic data, ahead of the reported numbers.

These are the "consensus" economic numbers which we cite every day.

Guzman has a problem with that "consensus" and the way economic forecasting is done; she doesn't think it's very accurate.

She has strong credentials. She has a PhD in Finance and Economics and worked for 17 years with Nobel Prize winner Lawrence Klein, a pioneer of modern economic forecasting. She was also a research assistant to Nobel Prize-winner Joseph Stiglitz.

Her beef: traditional economics is based on assumptions, not on data. "There's no science in the dismal science," she laments.

Her solution calls for using technology to study what people really are doing, rather than what economists assume they are doing. That, she says, will enable much more accurate forecasting.

On her website, subscribers can pull up a dashboard of roughly 4,000 indicators on virtually every kind of economic activity. The website provides tables of all the predictions and a graph of the predictions placed over the actual numbers.

And — this is the cool part — for some of the indicators you can watch the predictions in real time, part of the program dubbed LiveWire. The data is updated continuously, by the second.

Think about that. The Fed updates its estimates once a quarter, and most federal economic data comes out once a month.

Stare at the Consumer Price Index LiveWire, for example, which is sliced into 14 separate pieces, and after a few minutes the screen will suddenly flash yellow, and the numbers--estimates for the September report, out tomorrow--will change, reflecting an update in the data. Red arrows will appear indicating the estimates are lower, green indicating they are higher.

Right now there are three primary economic indicators offered in live time: consumer price index, personal income, and producer price index, though all three have dozens of sub-indices.

She will be rolling out retail sales shortly, with more to come.

OK, so what's the secret sauce? How is this information gathered and updated? Guzman is reluctant to get into a discussion about the details of her methodology, but she summarizes it thus: "It's the wisdom of crowds, but with a whole lot of math."

When I asked for more specifics, Guzman notes that the Internet is the perfect vehicle for exploring what people are really worried about and how they really feel, rather than what they say they are worried about or feel.

A paper she published that studied inflation concluded that people are far more likely to do Internet searches for "inflation" when they are worried that prices are rising, which makes sense. But she argues that this can be used to predict inflation expectations more accurately than surveys.

That's the wisdom of crowds part. The next step: she claims to have developed sophisticated algorithms that quantify those behaviors.

And how does she feel about the math? She says she is beating the Wall Street consensus.

Based on a brief look at her recent estimates, she seems to have had some hits. I was on the NYSE floor on the morning of Sept. 30 when the Chicago PMI came out. A number of traders had obtained a "private" (not published) forecast of 48.7 from Now-Cast. The consensus was for 53.4, so Now-Cast's estimate was a real outlier.

Traders were eager to see how close her estimate was. The reported number: 48.7, right on the nose.

August personal income was out Sept. 28. Consensus was for a gain of 0.4%; the Now-Cast estimate was for a gain of 0.3%. The actual number: a gain of 0.3%.

The Consumer Price Index (CPI) for September will be out tomorrow. Consensus is for a month-over-month DECLINE of 0.2%, Now-Cast's estimate is the same: a decline of 0.2%.

But that will keep changing, until midnight tonight.

Here's what I would like to see: an analysis of her predictions, against the predictions of the top strategists on Wall Street.

I certainly agree that Wall Street needs to improve its data analytics. What we need to determine is whether Now-Cast's methodology is clearly superior.

Programming note: Guzman will visit with Pisani on Power Lunch Wednesday, at 1:30 p.m., ET.

  • Bob Pisani

    A CNBC reporter since 1990, Bob Pisani covers Wall Street from the floor of the New York Stock Exchange.

Wall Street