Wall Street Analysts Need to Throw Away Their Models

By John Banks

It almost feels inhumane to be focused on markets right now.

After all, the United States is headed into its worst crisis in 75 years (since World War II), if not the Great Depression (1930s) or the Spanish Flu aftermath of World War I (1918).

And yet, retirement nest eggs still matter, and savings still matter. So, the stock market still matters, too.

Pandemic realities aside, we’re concerned by growing calls for a stock market bottom. The proliferation of Wall Street analysts saying “now is the time to buy stocks” has us concerned. 

And the big rebound that markets saw last week — the Dow’s most powerful surge since the 1930s — was actually a harbinger of 1930s-style markets in our view, defined by +20% roller coaster moves in both directions.

Moves like that are not a reason to be bullish. They are confirmation of a 1930s echo. This is, if anything, a reason to be incredibly vigilant (and prepared to weather a lot more downside). 

In our view, the S&P 500 has not fallen enough. The true “bottom” is nowhere in sight.

We feel the urge to pound the table on this because so many are pounding the table in the opposite direction. We fear they are dead wrong, in a dangerous moment for investor portfolios.

A big part of the problem, in our view, is the models these analysts are using. Their models were built for a world that no longer exists.

This is true even of the “quants,” the analysts who crunch numbers and pay no attention to headlines.

The quants have also entered a modeling “no man’s land,” in which most of the lessons they have learned and most of the assumptions built into their models should logically be thrown away.

The problem is embedded in the data set.

All of the data gathered over the past 30 to 40 years or so — from the mid-1980s to 2019 — implicitly assumes that market behavior in the future will look like market behavior in the past.

But that assumption is no longer valid, because we are facing a new kind of crisis now, of a type the markets have not seen in decades.

From about 1987 onward, every stock market crisis has more or less been a financial crisis.

In a financial crisis, there is a generalized pattern that looks like this:

  • There is a big dislocation involving the financial system.
  • The central bank steps in with rate cuts or stimulus.
  • New liquidity helps the crisis resolve quickly.
  • The markets carry on as before.

This is the pattern that’s held every time from 1987 onward.

Then, too, the tech bubble of 2000 was a relatively small slice of the economy, and the 2008 global financial crisis was mostly met with financial medicine.

The real economy — meaning Main Street rather than Wall Street — took a hard hit after 2008, but it didn’t topple into a full-blown crisis for Main Street, so the authorities ignored it.

The problem now is that all of the models these analysts are using — from the numbers in their spreadsheets to the rules of thumb in their heads — are based on a 30-year string of rising markets punctuated by quick financialized recoveries.

These analysts don’t have a “real-economy” crisis — as opposed to a financial crisis — living anywhere within their 30-year data sets. That means all of their crisis assumptions are wrong. Consider why the following statement is useless:

“The last four times the stock market dropped sharply, it recovered in X period of time.”

It is useless because the data set that covers “the last four times” is populated with instances of financial crisis. Again, there is no real-economy crisis in the data set, and that is what we are facing now.

Then, too, when the real economy breaks in ways that nobody understands — because nobody has ever seen anything like the current situation before — outcomes are unknowable by definition, in ways that offer nothing but downside.

On April 1, for example, tens of billions of dollars in rent went unpaid.

Imagine being the owner of a middle-income apartment complex and seeing 70% of your cash flow disappear. Or imagine owning a shopping center and seeing 70% of your tenants declare chapter 11 in one swoop.

Statistically speaking, that is never supposed to happen. It breaks all the statistical probability tables. Yet it is happening now, right now, all over the country. 

How are we supposed to “bounce back quickly” from something like that on a national scale?

And how is a 33% peak-to-trough stock-market decline, in the face of a dislocation that serious (and still ongoing with many ways to get worse), supposed to be enough?

Those questions are rhetorical because they have no good answer.

The simple reality is that analyst models forecasting a bounce back are rooted in the past — anchored to data sets that no longer apply — with a lack of accounting for the devastating factors of a crisis in the real economy here and now.

We get why analysts are doing this. They’ve been using their historical models and quant-based data relationships for their entire careers. It’s what they know. It’s what their “gut feel” is based on.

Their reactions and assumptions honed over 30-odd years of bull market movement, give or take a relatively quick flash crash and financial crisis or two, feel tried and true.

Also, they want markets to go up again (because this is what their clients want). They want the world to feel familiar. And they want to feel comfortable.

But none of that is justification to detach from reality, or to rely on data sets that no longer apply.

High-quality analysis should reflect the fact that we are dealing with a totally different animal now — a crisis in the real economy with a 100-year pandemic alongside. (And an energy sector meltdown to boot.)

Historical knowledge of markets will help, but not a myopic focus on the history of the past 10 years, or even the past 30 years.

Instead it will be increasingly necessary to look further and further back for potential parallels — to the 1970s, the 1930s, and the 1920s, for example, if not yet further back to events like the 19th century railroad boom or the panic of 1819.

It’s time to live in the present. It’s time to throw the old models away. It’s time to piece together the way forward from real-time analysis of fast-moving information variables in the here and now, instead of leaning on invalidated data sets that fail to apply.

Like it or not, it’s a new world now. The past is gone, forever.