A simple application of Bayes’ theorem can provide insights into otherwise secret understandings. A good example is the Shanghai Accord. This was the understanding reached among the United States, China, Japan, and the Eurozone on the sidelines of the G20 meeting of finance ministers and central banks in Shanghai on February 26, 2016. Those four G20 members comprise two-thirds of global GDP and make up a de facto G4 inside the G20.
The problem confronted by the G4 in Shanghai was that growth in China and the United States was slowing dangerously, and global growth was weakened by that slowdown. Structural reforms were stalled by political gridlock. Fiscal policy was constrained by already excessive debt. Monetary policy was increasingly ineffective, even counterproductive. With structural reform, fiscal stimulus, and monetary ease off the table, the only stimulus channel left was a return to the currency wars.
A cheaper yuan gives a temporary lift to China even if it comes at the expense of its trading partners. China devalued unilaterally in August and December 2015. Both times U.S. stock markets crashed in the aftermath. The G4 needed to find a way to cheapen the yuan without destabilizing the U.S. stock market.
The solution was to maintain the peg between the yuan and the dollar, then devalue the dollar. The yuan gets cheaper relative to the euro and yen, while the yuan-dollar peg is unchanged.
This meant that Japan and Europe would suffer a stronger currency and a trade disadvantage. That’s how currency wars work. For every winner, in this case China and the United States, there are losers, in this case Japan and Europe. A cheap yen had prevailed since 2013, and a cheap euro since 2014. Japan failed to make needed structural reforms. Now it was out of time. A new cheap-yuan, cheap-dollar phase was about to commence. The world’s two largest economies—China and the United States—needed help. This was the essence of the Shanghai Accord.
The challenge for analysts is that initially there was not a shred of evidence to prove the accord. The G4 meeting was conducted in secret and no explicit press releases or other information was shared. Analysts scoffed at the idea of a Shanghai Accord. Prominent foreign exchange expert Marc Chandler of Brown Brothers Harriman, writing about the Shanghai Accord, said, “Conspiracy theories have run amok.”
Bayes’ theorem allows an analyst to do better than conspiracy theories. A geopolitical action like the Shanghai Accord with scarce hard data to confirm it is the type of event Bayes’ theorem is designed to validate. The process is like a detective solving a crime with no witnesses. You collect evidence and interview suspects until you have a solid case.
To illustrate, consider ten discrete events in a row. Each event has a binary outcome: two possible results that tend to prove or disprove a starting hypothesis. Consider the binary outcomes as “heads or tails.”
These binary outcome events have two types. The first type is random. This is like tossing a coin. You could get heads or tails with equal probability, yet you never know in advance which it will be. The outcome of each coin toss is independent of prior tosses. The second type is path dependent. This means each event depends on prior events or relates to a single determining event.
If the Shanghai Accord hypothesis is true, relevant events would be path dependent. Central bank decisions would all be affected by the secret deal. Policy would not be a random coin toss. Events would be affected to some extent by the secret understanding.
The next step is to look at actions by central banks and consider what outcomes one would expect to observe if the Shanghai Accord hypothesis is or is not true.
If one tosses coins, what are the odds of ten heads in a row? Each coin toss has a 50 percent chance of being heads, and no coin toss is affected by another. The odds of ten heads in a row are roughly one in one thousand. (Mathematically this is expressed as (1/2)10. This can be put as: 0.5 × 0.5 × 0.5 × 0.5 × 0.5 × 0.5 × 0.5 × 0.5 × 0.5 × 0.5 = 0.0009765625 ≈ 0.001.)
One chance in a thousand is not impossible. If a daily possibility, it occurs about once every three years. Still, the odds against it are extremely high. No investor would base a trading decision on a sequence of ten heads in a row even though it cannot be ruled out.
Now, consider ten critical events that actually happened between February 26 and April 15, 2016. Each event had a binary outcome in advance. Call those that confirm the Shanghai Accord heads, and those that refute the Shanghai Accord tails. Reserve judgment for now on whether these events are random or path dependent.
Here are the events:
February 26, 2016: Before the G20 meeting is quite over, Fed governor Lael Brainard gives a speech in New York and says, “It is natural to consider whether coordination can improve outcomes . . . cooperation can be quite helpful.” Heads.
February 27, 2016: At the conclusion of the Shanghai G20 meeting, U.S. treasury secretary Jack Lew says, “We’ll keep each other informed . . . we’ll avoid surprising each other.” Heads.
February 27, 2016: Also at the Shanghai G20 meeting, IMF managing director Christine Lagarde says, “There was in the room a renewed sense of urgency, and a renewed sense for collective action.” Heads.
March 10, 2016: The ECB tightens policy relative to expectations by announcing that it has no plans for further easing. Heads.
March 15, 2016: The Bank of Japan tightens policy relative to expectations by not expanding its program of quantitative and qualitative easing. Heads.
March 16, 2016: The Federal Reserve eases policy relative to expectations by striking a dovish tone in its press conference. Heads.
March 29, 2016: Federal Reserve chair Janet Yellen makes her new dovish policy explicit with a speech at the Economic Club of New York. Heads.
April 13, 2016: Luc Everaert, the IMF’s mission chief for Japan, with reference to market intervention to weaken the yen says, “There is no good reason for Japan to intervene at this point.” Heads.
April 14, 2016: Christine Lagarde warns Japan that the IMF’s conditions for foreign exchange intervention to weaken the yen have not been met. Lagarde also says she is “quite pleased” the Fed moved to a dovish stance based on “the international status of the economy.” Heads.
April 15, 2016: An unnamed ECB official tells Reuters, “There was an in-principle agreement about exchange rates expressed at the G20 communiqué.” Heads.
There are additional data points, yet this list is sufficient to draw conclusions.
What does the sequence above represent? Did one randomly toss coins and observe ten heads in a row with 1,000-to-1 odds against? Or did one see exactly what would be expected if the Shanghai Accord were true?
It is highly probable that this sequence is not random, but path dependent. These later events all depend on one initial event—the secret Shanghai Accord.
Importantly, it was not necessary to wait until April 15, 2016, the end of this time line, to draw conclusions. The hypothesis was formed on February 26 based on official remarks at the close of the G20 meeting and Brainard’s speech. Subsequent data validated the hypothesis, but were not needed to create it. The hypothesis simply got stronger as time elapsed based on conditional probabilities.
By using Bayes’ theorem, an investor could pursue winning strategies—long euro, long yen, long gold, short dollar—with confidence, while Wall Street was still bemoaning “conspiracy theories.” It’s a matter of taking the math from New Mexico to the marketplace.
Complexity
Bayesian technique is not a science in itself, it’s an applied mathematical tool with robust predictive properties. The prime science of capital markets is complexity theory.
Capital markets are complex systems, yet complexity is little understood and even less used in financial economics. From the 1998 global liquidity crisis, to the 2000 tech bubble collapse, to the 2008 panic, policymakers have led the world into one crash after another. Their failure to use complexity theory explains why.
The case for complexity theory is straightforward. It’s not difficult to grasp. Investors must grasp it now if they wish to preserve wealth. The next panic will be too late. The ice-nine solution will lock down wealth and make it impossible to take defensive measures.
Complex systems have existed since the beginning of time. The creation of the universe in the Big Bang more than 13 billion years ago instantaneously led to complex dynamics in stars, gases, galaxies, and finally planets. What is new is our understanding of complexity as a formal science. That understanding is traced to the 1960 Lorenz experiments.
The timing of Lorenz’s breakthrough was not a coincidence. Prior to 1960, large-scale computing power was available to relatively few scientists, and was mostly applied to traditional problems in physics and operations research. Personal computers were still decades away. Yet by 1960, time-sharing on mainframe computers was available to researchers in more diverse fields, including Lorenz’s field, meteorology.
Without computing power, complex dynamic system paths were impossible to observe in graphical form. Humans could see complex system outcomes in tsunamis, fires, and floods. Still, they could not see the dynamics. Computers changed that.
To see what complexity is, it’s useful to know what it is not. Many systems are complicated, yet not necessarily complex. A handmade Swiss watch is complicated, but does not produce the highly unexpected behavior associated with complex systems.
Everyday phenomena such as a coin toss, a roll of the dice, or the spin of a roulette wheel are not complex phenomena. Such random processes are highly predictable. You don’t know if the next coin toss is heads or tails. You do know if you toss a coin a thousand times, you are certain to get about five hundred heads and five hundred tails. The odds of getting nine hundred heads and one hundred tails are so infinitesimal as to be effectively zero.
Also, random processes such as coin tosses and throws of dice have no memory. This means that a prior coin toss has no impact whatsoever on the next one. Some gamblers who see three heads in a row assume the odds are good the next toss is tails. This is called the gambler’s fallacy because it rests on a false assumption. The odds on each coin toss are always fifty-fifty. This is why a thousand tosses produce about five hundred heads and five hundred tails even if smaller samples produce an occasional run of heads or tails. When a short-run skew develops, you can be certain that future coin tosses will move the overall distribution back toward fifty-fifty, a phenomenon known as mean reversion.
Complex systems in contrast are highly unpredictable. A complex system can produce unexpected results seemingly from nowhere. In capital markets, phenomena such as market crashes, panics, and sequential bank failures are examples of complexity in action.
What is complexity? How can investors use an understanding of complexity to preserve wealth?
Complex systems are everywhere; they are not confined to laboratories or subatomic structures. If you drive a certain road every day that is uncrowded, then one day run into a traffic jam for no apparent reason, you have encountered complexity in action. Deciding if your favorite restaurant will be too crowded on a Friday night or whether the stock market is a bubble is an exercise in solving complex problems. Complexity is ubiquitous.
Complex systems are natural, man-made, or combinations of the two. A hurricane is a natural complex system. A stock market is a man-made complex system. A nuclear explosion is a combination because the natural complexity of uranium atoms is engineered by scientists to the supercritical state that releases a bomb’s destructive power.
Complexity theory begins with two tools. The first is the agent. An agent is simply an actor in a system. An agent can be a human in the case of capital markets or an atom in the case of a bomb. The agent is the irreducible unit generating behavior behind complex dynamics.
The second tool is feedback. This means that initial behavior produces output that affects subsequent behavior. This is why complex systems are said to have memory. When agents act in complex systems they observe prior acts that condition what they do next. Another name for the same idea is adaptive behavior. An agent adapts her next move based on what she has learned from prior moves in the system.
Random systems such as coin tosses, dice, or roulette wheels do not contain feedback loops. A coin does not adapt its behavior. In complex systems, behavior adapts all the time. Adaptation is one reason complexity produces surprising results.
Feedback is endogenous or exogenous. Endogenous feedback is internal to an agent, a matter of learning from mistakes. A cat that jumps on a hot stove learns not to do it again. Exogenous feedback is external to an agent. For a stock trader, it’s a matter of watching the behavior of others distilled through market prices. Markets may be going up, down, or nowhere, yet one observes this behavior before making the next move.
Agents and feedback are the building blocks of complex systems. What else is needed? It helps if the agents are diverse. If the agents are identical, feedback is weak because one agent’s behavior reinforces others’ behavior rather than changing it. In stock markets, diverse agents exist as bulls and bears, longs and shorts, rich and poor, old and young. Diversity of agents in capital markets is strong.
Another requirement is that agents communicate and interact in some way. Diverse agents do not give rise to complex behavior if agents cannot connect. Fifty cavemen sitting in fifty caves may have diverse views about the best way to hunt for food. Still, if they don’t leave their caves and communicate, diversity doesn’t matter. It’s only when cavemen leave their caves, gather around a fire, and start to share ideas that complex behavior emerges.
As diverse agents interact, adaptation begins. Once cavemen start to compare notes around the fire, some change their hunting method based on others’ success. Cavemen who don’t adapt may starve. A society of more efficient hunters starts to emerge. This is bad news for the mastodon, but good news for the cavemen.
Instead of cavemen, imagine a much larger group of stock traders hunting for the best trades. They have diverse opinions. They communicate on Bloomberg, Reuters, email, and the Web. Interaction is measured in trillions of dollars of daily trading volume. If a portfolio is losing money, the adviser needs to adapt quickly. You learn from others; others learn from you. Those who don’t adapt lose clients, or lose their jobs. They are soon out of the game. In short, capital markets exhibit all the characteristics of complex systems in a strong form.
These building blocks of complex systems are easy to understand. You need autonomous agents with diverse qualities. The agents need communication channels to interact. The interaction produces new information that feeds back to the agents. Then agents adapt their behavior to improve outcomes in the future.
Complex models do not resemble stochastic models used by central bankers. They do resemble the real world.
Feedback
Complexity in capital markets can be limned in social terms. Is there hard empirical evidence to support the view that capital markets are complex systems? Are there repeatable experiments that prove the point in accord with formal scientific method? The answer to both is yes.
Adaptive behavior arises in many socially based complex systems such as markets, traffic flows, and dating. The source of adaptation comes from competition for scarce resources. If valuable resources are available in unlimited quantities, one does not need survival strategies or adaptive behavior. You just take what you want. Scarcity is what causes individuals to adopt strategies to secure their share of the resources. This problem of allocating scarce resources is the foundation of economic science.
In capital markets, the scarce resource is wealth. In traffic, the scarce resource is a fast lane or a parking space. In dating, the scarce resource is the ideal mate. When competing for scarce resources, you need to make smart choices that increase your chances of winning a highly competitive game. If your trades are losing money, you can’t find a parking space, or y
ou can’t get a date, it pays to look around and see what the winners are doing. This is adaptive behavior.
One example is Warren Buffett, a winner when it comes to wealth. The SEC requires a quarterly portfolio disclosure by Buffett’s company, Berkshire Hathaway. When investors get a look at what Warren Buffett is doing, they copy his trades in the hope of becoming winners too.
This behavior results in the formation of a crowd whose behavior reinforces others’ behavior in the same crowd. The problem over time is that the winning strategy gets too crowded and no longer works. The first hipster to find a newly opened bar in Brooklyn with great live music may enjoy a few blissful weekends there. Eventually word gets around, the bar becomes crowded, and the hipster has to fight to get a drink. The winning strategy of hanging out in a cool bar becomes a losing strategy of standing-room only. The hipster moves on. So does Buffett.
This adaptive behavior displays feedback and memory. If you recall the bar as cool and uncrowded, you’ll go back. If you recall the bar as crowded and noisy, you might not (although some like noisy crowded bars).
To analyze crowds, physicists posit the formation of anticrowds. An anticrowd attracts followers that do the opposite of the original crowd. Such crowd-anticrowd behavior exhibits lots of memory and feedback. What separates the crowd and anticrowd are their expectations.
Using the bar example, it’s the case that some nights the bar is crowded and some nights there are empty tables. You just don’t know in advance. Agents make a forecast based on the best information available. Information might include social media updates from friends at the bar. Real-time information accelerates an agent’s reaction function, but it doesn’t negate it.
People considering going to the bar, or investors deciding whether to buy a particular stock, fall into three groups for forecasting purposes. The crowd believes the future will resemble the past. The anticrowd believes the future will not resemble the past. The third group does not have a forecast, but mentally tosses a coin and acts based on the random outcome.
The Road to Ruin Page 12
