Artificial Intelligence in the Stock Markets

Artificial intelligence has been much in the news recently.  Well-recognized deep thinkers have propounded profound thoughts that predict good and bad outcomes for artificial intelligence as it becomes ever more of a reality.  The takeover of the world by machines is inevitable--with monstrous consequences--or not, depending on who you ask.

There isn't much real-world empirical data relevant to the opposing sides of this argument.  But one big clinical trial is underway:  in the stock markets.  Most of the trading done today in the stock markets consists of computerized trading.  Institutional investors, like mutual funds, pension funds, and so on, comprise most of the rest.  Mom and Pop, trying to invest a few nickels for their retirement, are like pedestrians surrounded by massive semis barreling along at interstate speeds.

Much of the computerized trading is done by dynamic computer programs.  In other words, the computer doesn't buy and sell based on a static algorithm embodied in the program coding.  The program can change itself in response to market conditions and activities.  In essence, depending on what the program detects is happening in the market, it can alter its own coding without the need for a human programmer to keypunch and proofread line after tedious line of code.  The details of how these dynamic programs work are generally shrouded in commercial secrecy.  But we have a situation where computer programs take note of what's going on in their working environments, think about how to change themselves to be more effective (i.e., profitable) in light of changing conditions, and then alter themselves to do better.  That's getting rather close to what people do: try to change and improve themselves in order to advance their careers or make more money in their professional activities.

We have seen in recent years how computerized trading can cause mini-crashes and other short term  turbulence in the stock markets.  Mom and Pop are often finding the waters too rough, and suddenly see the virtue in the paltry returns of passbook savings accounts or certificates of deposit whose yields have been flattened by the nearly supine yield curve.  Even some professional money managers are looking for ways to fly over or around the storm clouds of computerized trading, moving trading to venues that claim to allow only "natural" (non-computerized) investors to participate.

For academic researchers and other prognosticators, the dynamic computerized trading in the stock markets could furnish a useful body of data with which to work.  The rest of us, unwilling guinea pigs in a clinical trial we didn't sign up for, can only look to financial regulators and other government officials to ensure that the results of the experiment don't turn out too badly,

Are Stock Prices Real Any More?

One wonders if stock prices are real.  Consider the evidence.  Due to political obtuseness, the U.S. barely avoided defaulting on its debt.  The economy continues to expand at a disappointing rate.  Employment growth is tepid.  Middle class incomes are falling, on average.  Consumers are gloomy.  The Federal Reserve Board is gloomier.  Businesses hold back on investing cash.  But, last week, the S&P 500 reached record heights.  When you encounter cognitive dissonance in the financial markets, be careful.  Every time in the past 20 years when things seemed out of whack, it eventually turned out that, in fact, they were out of whack.  In other words, if it looks too good to be true, it probably isn't true.

Could investors be wearing their stupid hats again?  Their 401(k)'s got clobbered in the 2000-01 tech stock crash, and after adjustment for inflation, stocks still haven't recovered.  Investors' 401(k)'s and homes got clobbered again in the financial crisis of 2008, and many haven't recovered from those losses.  What does it take to move up the learning curve?

Maybe, however, the problem isn't investors.  Over 50% of the trading volume in the stock markets comes from high-speed computerized trading.  This activity isn't based on human judgments.  It flows from algorithms and formulae.  Some of the computerized trading is dynamic--it changes based on what it observes in the market.  Since the activity it is often observing is computerized, we now have computers reacting to the activity of computers, which could be reactions to activity by other computers. 

When stocks were valued by humans, we had some idea of what we were dealing with.  Even if things seemed irrational or even bubbly, we could understand what was happening, albeit with a frown.  Now, with stocks being priced by "thought" processes that are impenetrable to the average investor, the market not longer reflects the collective judgment of humans.  Instead, it is an amalgamation of valuation processes that often aren't based on human judgment.  The things that people are concerned with--lousy economy, sluggish jobs growth, falling incomes, dim view of the future--may not be finding their way into stock valuations, at least not in the ways that they historically did.  If so, we can't be sure stock prices are real because we have no idea what the computers will do next.  Caveat emptor.

Why No Great Rotation?

A popular view among market aficionados is that, with bond prices falling while stocks have been rising, money would shift from bonds to stocks.  Stocks and bonds have historically often moved inversely.  When stocks rose, bonds fell, and vice versa. With bonds falling now, it would seem reasonable to expect investors to rotate their money into stocks.  But there has been no rotation.  Why not?

First, for the past five years, we've had a brave new Fed which has manipulated asset values in ways beyond historical experience.  Since early 2009, central bank easy money has helped to spur a stock rally accompanied by a bond rally.  Both asset classes rose simultaneously, instead of moving inversely.  With their traditional relationship out of whack, it is hardly surprising that they don't cha-cha when they're supposed to.  Investors would be understandably suspicious of stocks in a market that is seemingly dependent on the Fed's methadone program, especially when the Fed is talking about easing out of its role as Dr. Feelgood.

Second, the Great Rotation is an investment strategy for the medium to long term.  Today's stock market is dominated by high-speed, computerized trading, where the holding period for stocks is measured in milliseconds.  The long term human investors that might consider rotating greatly have mostly been supplanted, and many have chosen to invest on autopilot, buying index funds and throwing salt over their left shoulders.

So whither the markets?  That's the $64,000 question, and in truth nobody knows the answer.  With both stocks and bonds having enjoyed years-long bull markets, logic and experience, especially recent very painful experience, tell us that when markets can't keep rising indefinitely, they won't.

Knightmare in the Financial Markets

Knight Capital's announcement today that it lost $440 million yesterday (Wednesday, Aug. 1, 2012) when its trading system went haywire illustrates the potentially dramatic consequences of computerized trading. Details on exactly what happened remain scarce. But it appears that trading volume at Knight spiked for some 30 to 45 minutes at many strange prices. A large number of trades were cancelled. Broker-dealers that normally route orders to Knight have temporarily ceased to send it business until the air is cleared.

Knight opened for trading this morning, saying that it still complied with regulatory capital requirements. Nevertheless, its capital position is reportedly not pretty, and news stories indicate that it's seeking a capital infusion, plus an emergency loan from a major bank. Knight may not survive if it doesn't establish confidence among the broker-dealer community within a day or two.

And that's the really scary thing about the Knight debacle. It's conceivable that Knight could have been rendered immediately insolvent if its big trading glitch had lasted a bit longer, or had involved a somewhat larger number of transactions. Such an insolvency would have impacted Knight's counterparties, possibly imperiling them if their exposures to Knight were large enough. If these counterparties had become insolvent, the financial miasma could have spread and other firms knocked down like falling dominoes. All this, potentially, because of less than an hour's computerized trading gone haywire.

The risk of insolvency in such a situation could be heightened by the fact that other market participants might observe such a debacle in real time and pull their accounts before the trading day ended. This, were it to occur, would amount to a run on Knight. Now that the financial community knows that Knight (and perhaps its competitors) can become imperiled within an hour's time, they might be all the quicker to grab for their money first, and ask questions later. The quaint display of depositor anxiety so sentimentally portrayed in It's A Wonderful Life would be a mere box car compared to the Maserati of broker-dealer flight in our world of computerized trading.

Sadly, regulators would probably have little or no idea of what would be going on. The SEC recently adopted rules for a consolidated audit trail with a requirement to report trades to the agency. But the new rules call for next day reporting to the SEC. The agency wouldn't be able to track in real time what the cannoli was going on, and hence would be behind the curve as market participants cut and ran.

Of course, the Federal Reserve would jump in with bailout checks for one and all of the imperiled firms if a meltdown of the financial system loomed. But the availability of desperation-driven, last ditch bailouts offers little comfort. More than ever, market participants and regulators need to get a handle on computerized trading. The ability of computers to execute vast numbers of ridiculous trades without any constraint is really, truly, seriously dangerous. By all indications, the financial markets are now operating on a sudden death basis. That cannot end well.

The China GDP Head Fake?

Yesterday, April 12, 2012, rumors in the morning about China's first quarter GDP growth coming in around 9%, higher than the expected 8.4%, fueled a stock market rally. (See http://blogs.wsj.com/marketbeat/2012/04/12/stocks-jump-china-gdp-whisper-number-fuels-rally/). The Dow Jones Industrial Average bounded up 183 points.

Today, China's first quarter GDP growth was reported at 8.1%, lower than expectations by 0.3%. (See http://money.cnn.com/2012/04/12/news/economy/china-gdp/). The Dow closed down 136. The rumor was wrong, seriously wrong.

Maybe it was all an innocent mistake. After all, today's Friday the 13th. Maybe someone just misinterpreted something and spread the misinterpretation. But one thing's for sure. Some people made a bunch of money trading the market up yesterday, and then trading it down today. Volatility makes money for Wall Street pros, including market makers, specialists, high speed traders and so on. But some less cynical market participants, who bought and held overnight, probably lost money and quickly.

Volatility can come from exogenous sources. The idiocy underlying the EU's structural problems and its sovereign debt crisis weren't creations of Wall Street. But they sure as pumpernickel have caused a lot of volatility.

There's also home grown volatility, emanating from Wall Street sources that might have a good day at the office if the market is hopping. Spreading truthful and accurate news is generally okay, unless it happens within the context of insider trading. But spreading false information can make regulatory brows furrow.

It's difficult to investigate rumor mongering, especially if the rumor concerns aggregate economic data (as opposed to company-specific or security-specific information). But trading surges triggered by false information undermine investor confidence. Natural investors (i.e., those that buy with the hope of profiting from investing, as opposed to make a quick buck from trading) are the foundation of the financial markets. But, with the 2000 tech cash, the 2008 financial crisis, and the 2010 flash crash, they are leaning, if not running, toward the exits. It really doesn't help when they are jerked around by false rumors.

Financial Regulatory Reform: Ask Basic Questions About Computerized Trading

One of the principal reasons for the financial crisis of 2007-08 was the failure--of Wall Street executives, lenders, investors and regulators--to ask basic questions about the computer modeling that was used to construct the now infamous derivatives contracts whose flaws triggered our lengthy, ongoing recession. While these models covered, more or less, 99% of the potential market scenarios that could arise, they didn't cover the possibility that the real estate market might, on a national level, fall. Such a phenomenon had not occurred in a very long time. And because of that, the big players in the derivatives markets presumed it wouldn't occur. That was a mistake. We are still paying the price.

High speed computerized trading has recently been in the news. Outsized profits at Goldman Sachs, followed by outsized bonuses, were partly the result of electronic trading at speeds measured in the millionths of seconds. The alleged theft of some of Goldman's ultra-fast trading software by a former IT employee added tabloid sensationalism to the story. Complaints about flash trading, where some firms with superfast computers can see and trade ahead of other large orders, have led the SEC to consider banning the practice. "Naked access," which is about money, not sex (remember, this is Wall Street), involves a favored customer of a brokerage firm using the broker's identity to trade directly on an exchange. This would presumably heighten the speed of its trade executions. The high speed firm would, to the rest of the market, appear to be an exchange member. But it wouldn't necessarily behave like an exchange member and might destabilize the market. A third kind of high speed trading involves the arbitrage of prices in more than one market, an established trading strategy now conducted at exponentially faster speeds.

Computerized trading is now reported to be more than half the total trading volume in the stock markets. This represents a fundamental change. Back in the days when landing a human being on the Moon was still a dream, the stock markets consisted of the interactions of human judgments. Human interactions can be strange and even destructive. The history of financial booms, bubbles and busts demonstrates that. But at least people more or less understand each other most of the time. Family bonds, society and commerce wouldn't be possible if that weren't true.

But computer models are another thing. They don't always work as intended, because of imperfections in design by their human financial engineers. Yet, the image of computers as more reliable and thorough than humans leads to the notion that they are less prone to errors. That, as we know from the recent mortgage-backed mess, is demonstrably untrue. We need to know if computerized trading could have unintended consequences at a time when we really don't need more unintended consequences.

It's important to consider basic questions. Flash trading could create a two-tiered market, with better prices for well-computerized big players and crumbs for everyone else. This ultra-short term profiteering isn't in keeping with the baseline purpose of the stock markets--to provide capital for long term investment. Flash trading doesn't offer much in the way of societal benefits. Yes, there are arguments that it increases liquidity. But let's remember that not all liquidity is the same. There is the liquidity provided by exchange specialists and market makers, which is "counter-cyclical." This means that they sell when investors want to buy and buy when investors want to sell. Counter-cyclical liquidity is good for markets because it keeps things flowing. But wouldn't flash trading liquidity more likely be "pro-cyclical," meaning it would be used to buy ahead of a market surge and sell ahead of a market drop? This would destabilize markets.

It's also important to look at the collective interaction of these high speed trading strategies. When more than half the trading volume in the markets is computerized, it's fair to conclude that machines are observing and interacting with other machines a lot of the time. Perhaps flash trading or naked access could create price anomalies that a multi-market arbitrage system would believe to be the beginning of a profit opportunity. But, without the human element that made historic trading patterns what they were, the markets might move in unexpected directions. The hyper vigilance that a computer can provide might result in large wolfpacks of high speed traders pouncing on pricing anomalies with ultrafast trades, pushing prices far into the opposite direction and producing asymmetries never before seen. That might be bad.

And there's also the problem of inputs. Remember what they say about computers: garbage in, garbage out. Interest rates, for example, have historic relationships with the values of other assets. The Federal Reserve has pushed interest rates down to extraordinarily low levels. It's possible that the computers are misinterpreting what the human action by the Fed means. Perhaps these extremely low interest rates make the computers think that stocks are more valuable than they really are. The humans pretty much know that the Fed can't keep interest rates down with the flounders forever. But who knows when the humans at the Fed will begin the process of draining the vast oceans of liquidity they have dumped into the economy, how fast they will do it, and what mechanisms they will use. The computers may have contributed heavily to the upside delirium we have recently seen in the markets, where there is no such thing as bad news and all news makes the market rise. Interest rates--especially short term rates--are the product of government policy, and today we have government policies and intervention never before seen. Might the computers, programmed to react to historical patterns, misinterpret what's going on? If the computers exaggerate the upward momentum, might they draw in humans who--thinking they're seeing, not computers, but other humans buying--invest when they would otherwise be worried about jumping in at the top of an asset bubble?

There are surely many other questions that could be asked, especially if one can get a detailed understanding of how these computerized trading strategies work. The regulators can get that information, and should. It won't be easy to evaluate this stuff. A lot of it involves technical complexity and serious math, things that the liberal arts majors who heavily populate the legal staffs of regulatory agencies don't hanker for. But you can't regulate an industry if you don't understand it, and it's time to get a thorough understanding of what high-speed trading is all about, before unintended consequences once again wreak havoc in the markets.