Wednesday, January 27, 2010

Disaster Prediction, Trading Strategies, the Fall of Rome, and Volcanic Sunsets



(Turner, the Fighting 'Téméraire' tugged to her last Berth to be broken up)


"It was decided by the university of Coimbre that the sight of several persons being slowly burned in great ceremony is an infallible secret for preventing earthquakes."
- Voltaire, Candide


Predicting Earthquakes: Discouraging Results

Unlike volcanic eruptions, which do normally provide warning signs prior to becoming violent, earthquakes offer little in the way of forecasting aids. According to the "Self-Organizing Criticality" theory described in the previous post, the initial difference between the cause of a small earthquake and a large one may be surprisingly minor and technical. The problem is that while it is understood that seismic waves nucleate at the focus of the earthquake, the microstructure of the 'quake involves sharp, sudden events that occur without warning.

As a result, much of the work on earthquake prediction has attempted to find macro patterns in the data rather than to attempt to find ways to monitor the activities within the earth and determine when and where a major earthquake is going to take place. The goal is to find areas that may have achieved a critical stage and become sensitive, particularly if those areas are close to large human settlements.

As we know, earthquakes are statistically fractal, and follow a power law distribution. In any given year, there are about 1200 magnitude 5 earthquakes, 120 magnitude 6, 12 magnitude 7, and 1 magnitude 8. In other words, there is a magnitude 6 earthquake somewhere in the world every 3 days. The largest earthquake in a given year can release more energy than all of the other earthquakes that year combined. The central problems of earthquake prediction are that: 1) there is little in the way of a pattern of repeatable pre-quake behavior that could be used as a warning system, and 2) what limited evidence we do have does not allow us to distinguish between the pre-destruction activity of, say, a magnitude 8 earthquake and a magnitude 5 earthquake.

Earthquakes and Trading Strategies

For our purposes here, I will decompose the topic of "prediction" into three different levels of forecasting precision:

-Category 1: true prediction. This means being able to determine, in advance, when and where a major earthquake will take place.

The investment or trading equivalent would be able to state that a given company's stock price was undervalued and would increase to "fair value"---however that was determined---by a given date.

-Category 2: probabilistic edge. This means being able to gain a measurable performance advantage by being able to determine when and where a major earthquake is more likely to occur. Having a probabilistic edge would mean being able to commit a larger percentage of scarce resources to that area in advance.

The investment equivalent of this would be a generalized "market timing" approach that gradually deploys portfolio resources into, say, the S&P 500 and Hang Seng Index, and away from the Nikkei, because a forecasting model or valuation model has computed that there is an increased likelihood of the US and Chinese equity markets outperforming---relative to Japanese stocks---in the future (although a precise start date for this is unknown and errors could be large, the model tells you to be "overweight" the US and China markets and "underweight" Japan. It is essentially a resource-shifting, long-term allocation model, rather than a more dynamic approach that moves completely in and out of the markets on shorter time frames to try to capture discrete moves).

Category 3: agile reaction. This approach essentially gives up on 1 and 2 and tries to simply react to earthquakes as they occur, in real-time, as efficiently as possible. Efforts are made to decentralize operations in order to exploit opportunities (one of the tenets of Maneuver Warfare Theory), and to rehearse what the military calls immediate-action drills: pre-choreographed, codified response procedures---usually for emergencies and other truly decisive, time-sensitive situations---that have been designed to make the best use of information as soon as it emerges from the tactical environment.

The Evidence

In both earthquake and market prediction efforts, there is little evidence that Category 1---true prediction---is consistently possible. Indeed, an analyst can basically get one prediction right by sheer luck and then fall victim to post-dictive claims of having an extremely sophisticated approach (either because of a general vulnerability to the fundamental attribution error or because of a more cynical desire to capitalize on a lucky call and generate advisory fees from it). Geophysicists and geologists who study earthquakes professionally have basically given up on Category 1 prediction.

Category 2 fares a bit better. For example, we know that a major, seismically active subduction factory on the Pacific rim---the Indonesian archipelago, basically---has historically generated many of the most spectacular and dangerous earthquakes and volcanic eruptions. We do not know when, precisely where, or how big another natural disaster will strike the region, but we do have enough knowledge of the system to be able to predict that concerned parties should probably shift an above-average resource weighting to the region. This may sound intuitive, but it could actually save many lives to have disaster-relief assets forward-deployed in the region when another calamity occurs. The world is just too big to try to adequately cover disaster contingencies everywhere at the same time. As Frederick the Great warned: "By trying to defend everything, he defended nothing."

Category 3 is where much of the action in earthquake management can be found. The network of seismic detectors placed all over the world---as previously discussed, this was originally built for monitoring underground nuclear weapon tests---can give immediate data regarding a significant earthquake. How this data is operationalized will vary from region to region: for instance, tsunamis, or seismic sea waves, can be tracked in real-time and evacuation plans can be executed because tsunamis travel slow enough for it to do a lot of good (tsunamis move at about 500 miles per hour). When a tsunami has to cross a vast expanse of open ocean, the real-time tracking may give occupants of a targeted shoreline many hours to move to higher ground and prepare for flooding. In Japan, systems have been put in place to shut off gas valves, stop trains, and even send alert texts to individual cell phones when seismic detectors are tiggered, and companies regularly conduct drills for earthquake evacuation from buildings.

In terms of the financial markets, there is some compelling evidence that Category 3 methods---which are typically classified as "time-series forecasting" or "statistical extrapolation algorithms"---consistently outperform analysts who attempt to employ Category 1 and Category 2 prediction approaches. I will save this discussion for a future date.


Mean-Reversion and Trendfollowing in Earthquakes and Markets

A construction called elastic rebound theory puts forth the argument that stresses underground will build for relatively fixed periods of time and then will have to be released. Unfortunately, there is not a lot of evidence that earthquakes offer periodicities that can be used to gain a substantive edge. Similarly, studies of financial market periods---from annual "calendar effects" to 4-5 year business cycles to much longer Kondratieff Waves---have generally not uncovered applications of great practical use (with a few notable exceptions). However, there are some interesting ideas out there regarding the increased likelihood of earthquakes being more likely to occur at particular nodes along known fault lines.

For instance, the seismic gap hypothesis is a mean-reversion theory: it forecasts that the most likely place for a future earthquake is in a place along a fault that has not had an earthquake yet. There is a saying among traders that market "gaps get filled", and this is precisely what the seismic gap hypothesis argues. Such a gap analysis of, say, the San Andreas fault in California would reveal that the next major earthquake gap destined to be filled is, unfortunately, directly under the city of San Francisco.

A trader who employs this type of mean-reversion strategy bets that an elastic band connects the market's current price to its historical norm---if the price diverges too far from the norm, which is usually considered some kind of fair-value "equilibrium" price for the asset, a gap between current price and true value will form and the elastic band can be expected to yank it back. Thus, mean-reversion traders will tend to buy weakness and sell strength.

In contrast to the seismic gap hypothesis, other models of earthquake prediction suggest that the most likely place for future earthquakes is where earthquakes have tended to take place in the past: earthquakes in one location tend to be followed by more earthquakes in the same location, as the first earthquake may stress and weaken a fault area so that others can occur.

Similarly, a trader who employs a trendfollowing strategy bets that the term "historical norm" may be meaningless because markets evolve over time. Prices may diverge for any number of reasons, and may not return to a past, theoretical-equilibrium level for an extended period...if ever. Gaps may not get filled. Thus, these types of traders will tend to buy strength and sell weakness to position themselves within emerging trends.

The "earthquake trends" hypothesis may not, in fact, be directly opposed to the hypothesis of seismic gaps, because it is possible that they simply operate at different scales. Perhaps, in a larger-scale sense, major earthquakes do cluster, or trend, in certain regions of the world. It may be also true that, within those regions (i.e., at a more localized scale), earthquakes tend to fill local gaps. At an even high frequency scale, we can find trending behavior yet again---a major earthquake will produce aftershocks, or smaller earthquakes in the immediate vicinity, for days or even weeks.

There is no clear "winner" in the seismic gaps vs. trending prediction contest. Both sides can claim winners and losers.

We can similarly observe that both mean-reversion and trendfollowing trading strategies can be profitable in different macroeconomic environments. The two trading approaches have different return profiles---mean-reversion works most of the time and generates small winners, but has large and sudden losses when it stops working; trendfollowing generates small losses more than 50% of the time, but has large winners when it starts up again. Assuming that the mean-reversion strategies have proper risk controls in place and do not heavily engage in the martingale betting styles that have been previously discussed, both of these approaches can be quite profitable---can have "positive mathematical expectancy"---over adequate holding periods.

In much the same way as opposing earthquake prediction theories may perhaps be reconciled by adjusting the resolution to look at different scales, it appears that market trend approaches can make money at very short (seconds or minutes) and very long (weeks and months)time scales, while mean-reversion trading strategies may operate successfully at different frequencies.

Animals and Earthquake Detection






(can snakes, like this rather elegant King Cobra, predict earthquakes?)




There are anecdotal reports that some animals have the innate ability to predict earthquakes, and that they will begin behaving erratically prior to these and other natural disasters. The Chinese have probably done the most extensive research in this area (China has good reason to be motivated, given that earthquakes in China in 1920, 1927, 1932, and 1975 each killed over 100,000 people, and one in the 1500s killed over 800,000). To date, no evidence of a consistent earthquake prediction capability in wolves, snakes, chickens, and so on has been found.

It is possible, although of little practical utility for humans looking to escape earthquakes, that animals with higher sensory acuities may be able to detect P waves, which reach the surface from the earthquake foci before S waves do and are beyond the threshold of human sensory detection (seismometers detect them for us). It seems likely, given the highly irregular natural selection pressure posed by earthquakes, that any animal earthquake-detection ability would be a secondary benefit---a fortunate by-product---of another, more routinely useful, evolved system, perhaps one used for predation, escape, or navigation.

From the USGS: "Is it reasonable for a seismic-escape behavior pattern to evolve, and can such a genetic system be maintained in the face of selection pressures operating on the time scales of damaging seismic events? All animals instinctively respond to escape from predators and to preserve their lives. A wide variety of vertebrates already express 'early warning' behaviors that we understand for other types of events, so it’s possible that a seismic-escape response could have evolved from this already-existing genetic predisposal. An instinctive response following a P-wave seconds before a larger S wave is not a 'huge leap', so to speak, but what about other precursors that may occur days or weeks before an earthquake that we don’t yet know about? If in fact there are precursors to a significant earthquake that we have yet to learn about (such as ground tilting, groundwater changes, electrical or magnetic field variations), indeed it’s possible that some animals could sense these signals and connect the perception with an impending earthquake."


Development Economics and Natural Disaster Vulnerability

Besides the magnitude of the earthquake itself and the destructive energy that is released on the surface, the other reliable factor in assessing the casualty rates of quakes is the economic status of the region that is hit. Simply put, earthquakes hit the urban centers of developing economies very, very hard. The combination of weak building construction techniques and large populations makes for high casualty rates.

In California, to cite an example of a place where advanced earthquake-resistant construction technology is being employed, structures are basically bolted to their foundations (rather than just built on top of them). In contrast, a building just built on a free foundation slab---as is common in developing nations---can react very badly if hit by a high-amplitude Love or Rayleigh wave---the effect is something like what happens when a would-be prankster pulls a rug out from under the feet of his victim.

Another general rule seems to be that buildings constructed on bedrock fare better than those constructed on top of sedimentary deposits, which shift very violently during earthquakes. One of the reasons why a relatively low-power earthquake was able to wreak havoc in Mexico City in 1985 is because much of that city's metropolitan area was built on what is basically an ancient lakebed.

There are various advanced construction mechanisms that involve springs and buffer contraptions used to decouple the motion of a building from the motion of the ground during an earthquake, and advanced economies are experimenting with them. An intriguing problem presents itself because the damage that a building sustains in an earthquake may sometimes have more to do with frequencies than with magnitudes: a building that could withstand a magnitude 8 earthquake may collapse to a magnitude 7 if the building naturally resonates at the same frequency as the ground shakes do, and a self-reinforcing feedback loop of harmonic resonance occurs.

As stated in a previous blog post, earthquakes often cause uncontrollable fires: 140,000 people were burned to death when firestorms raged through Tokyo after the earthquake that hit Japan in 1923. Gas pipes are broken and power lines are brought down, causing fires to start, while at the same time the earthquakes attack water mains, disrupting fire-containment efforts. Developing economies may be at particular risk because fire and emergency services personnel may be chronically underfunded, flammable materials may be strewn everywhere, and fire codes may be nonexistent.

Earthquakes, Sanitation, and Disease

Even after the death and destruction caused by building collapse, fire, tsunami, and land slides have occurred, a developing region hit by an earthquake may face yet another public health crisis. Lack of clean water is a problem in developing regions during the best of times: you really see how water procurement needs affect the economic landscape when you travel to a relatively desolate region of the developing world. I can recall driving around in an NGO-marked Toyota Land Cruiser and touring some aid projects in remote areas of Tanzania with an AMREF mission. We would see women and girls walking for miles and miles on the sides of the dirt roads, forming long processions, to procure water and bring it back in the large, heavy-looking containers they balanced on their heads. This activity took up most of their days, every day, because the demand for water is unrelenting (when 50% of your workforce is unable to pursue education or employment because it is tasked with endless, manual water-transport duties, your economy will clearly suffer a productivity constraint).

Stated simply, refugee camps and forced migration facilities that are put in place after natural disasters such as earthquakes become virtual Petris dishes for infectious disease. Lack of sanitation, water, electricity, and sterile medical instruments, combined with packed living conditions, lead to outbreaks of everything from cholera, polio, and meningitis to HIV/AIDS and Ebola. Drinking supplies quickly become contaminated and it can be months before all of the contamination sources can be discovered. Once again, the higher population densities of urban areas makes for more difficult disaster relief efforts, as facilities that can offer adequate space, restrooms, and hospital beds simply may not be available.

(One possible component of humanitarian-relief efforts, at least in coastal regions, would be ships that are designated as dedicated "water factories"---basically mobile desalination and purification plants capable of producing large amounts of potable water from seawater, and then piping it to stations onshore for distribution via trucks or some other method. If the reader is interested in getting a great overview of the science, logistics, and politics of human waste removal, surely an unpleasant but critical aspect of public health, I recommend a book called The Big Necessity by Rose George).


The Catastrophe of David Keys






(Arnold Bocklin, The Plague)

The discussion of disease outbreaks following natural disasters is not limited to those illnesses that may result from poor sanitary conditions within refugee camps. One potential, quite serious source of contagion is unleashed when an earthquake or volcano disrupts a stable "pathogen reservoir", an area in which a potentially deadly microorganism has been contained by natural forces (usually the reservoir consists of animal hosts that are not hurt by the pathogen, presumably because of a symbiotic co-evolutionary relationship). A disruption to the habitat of the asymptomatic hosts and pathogens, the theory goes, can cause the plague to leave its stable zone and find a new home.

The most likely actor behind such a scenario will probably be a volcanic eruption rather than an earthquake, because a large eruption can release enough particulate matter into the upper atmosphere to cause global cooling.

Perhaps the most far-reaching example of a geologic event unleashing a plague on mankind is the grim story of the bubonic plague, the Black Death as it was known in 14th century Europe. In a provocative book titled Catastrophe, author David Keys speculates that a massive volcanic eruption in 535-536 AD (Krakatoa is ultimately implicated by Keys, although a volcano in Papua New Guinea has also been considered) put enough material into the upper atmosphere that global weather conditions were affected vis-a-vis "volcanic winter", as they were in 1815-1816 when Tambora exploded and caused crop failures as far away as New England (where it snowed in June).*

*(on the plus side, the paintings of Joseph Turner began to feature much prettier, more vivid sunsets after the Tambora blast, since the atmospheric changes created by the volcano had the side effect of increasingly the drama of sunsets all over the world. In fact, a 2007 study revealed that more than 500 paintings---from a portfolio that included Turner, Rembrandt, Reubens, and Degas---all featured far richer, more exciting sunsets in the years following volcanic eruptions).

Keys began his investigation when he learned that there was a pattern in tree rings, sampled all over the world, that revealed a sudden period of very low growth rates---trees virtually stopped growing, all over the world, all at the same time. Ice core deposits taken in both Greenland and Antarctica revealed much higher sulphate levels (a signature of volcanism), and these could also be traced to the 535 AD date.













(view across the Sunda Strait at sunset, looking at the infamous volcano implicated by Keys)


Besides leading to terrible famines as far away as Mexico (where the great city of Teotihuacan was ultimately abandoned), Keyes postulates that the cooling of temperature created instability in a "pathogen basin" in the Great Lakes region of Africa, ultimately leading to the spread of the bubonic plague---the "Plague of Justinian"---and the collapse of the Byzantine Empire that marked the final end of the Roman domination and the beginning of the Dark Ages.













(Hagia Sophia in modern Istanbul/former Constantinople, commissioned by Justinian)


Among his other justifications for an African origin for the bubonic plague (there are competing theories of origination), Keys found that the ivory trade in the great East African ivory port cities of the time period---Essina, Toniki, Opone, and Rhapta---all saw a drastic reduction in ivory production in the wake of the 535 event. Keys contends that the ivory centers would have been particularly hard-hit by the plague if the African ivory origin/distribution thesis is correct, and his research thus does lend support to his origin claim. The notion is that the plague would have spread from modern Zaire to Dar es Salaam or central Ethiopia to Essina, then up the eastern coast of Africa through the existing ivory trade route (the Romans had an insatiable appetite for ivory), landing at the major port of Alexandria, Egypt before entering the heart of the Byzantine Empire.

In Justinian's Flea, his book on the plague, William Rosen says: "From the moment humanity originated in East Africa, human populations in the origin basins of Tanzania and Ethiopia grew far more slowly than they would have anywhere else, because they were surrounded by the richest menagerie of pathological microorganisms on the planet, the evolutionary equivalent of a baby crib filled with deadly stuffed animals."

The bacterial demon that killed so many---Yersinia pestis---had previously found a nice home in the digestive tracts of various mammals. This limited its ability to move from host to host. Y. pestis later evolved in such a way that it could be carried by fleas; when the fleas withdrew blood from reservoir animals, Y. pestis was able to spread. When a flea bit an infected rat, the bacillus caused a clot to form in the flea's digestive system. The flea would be unable to pass the bacillus through its body, allowing the bacillus to multiply in its new host. Furthermore, the clot would cause the flea to become ravenously hungry.

It turns out that the fibrin clot that blocks the flea's GI tract does not seem to form in environments with temperatures above about 25 degrees centigrade. This is how the volcanic winter/plague connection works its black magic: by cooling temperatures all over the world (as evidenced by the tree ring growth patterns), the volcanic eruption allowed the clot-formation phenomenon to occur in the African fleas. The rats brought the fleas, insane with hunger, to human populations, and then the interspecies transfer was able to take place.

Keys argues that climate change caused by a huge volcanic eruption led to a movement of disease-carrying animals out of the stable reservoir area and into close contact with human communities. When the rats carrying the fleas that in turn carried the plague reached the Byzantine capital of Constantinople in 542, via the trade routes, not even Belisarius, Justinian's great master of warcraft and one of history's tragic figures, would have been able to win the battle that would come.

Before long, 5,000-10,000 people were dying every day. An estimated 100 million ultimately perished, the Roman and Persian empires---heavily dependent on long-distance trade routes that were abandoned because of contagion fears---collapsed, and the armies of Muhammad found little in the way of armed resistance as they left the Arabian desert.

..................................

Epilogue: Our old adversary, the Black Death, would return again to kill many, many more. About eight centuries later, global cooling of some origin---possibly another volcano---may have caused the Great Famines of 1315-1317, as well as floods along the Yellow River in China that drowned more than 7 million people. Bodies went unburied, the rat population dramatically increased, and rat-borne fleas carrying the bubonic plague made their way to the Middle East and Europe through the grain trading routes. An estimated 25 million people would be killed---about 1/3 of the population of Europe. Some areas would not recover their pre-plague population levels for another 600 years. To cite just one result, 50-70% of the population of Britain was killed off by the plague by 1400, a result that approximates the casualty estimates for a thermonuclear war. The 14th century is considered a cataclysm, among the most traumatic periods in all of human history.




No comments:

Post a Comment