Has High-Frequency Trading Destabilized Markets?
|March 27th, 2012, 06:33 PM||#1 (permalink)|
Aurora, Il USA
Futures Experience: Advanced
Favorite Futures: futures
Posts: 5,839 since Nov 2010
Thanks: 3,275 given, 3,321 received
Has High-Frequency Trading Destabilized Markets?
Over the last week, we’ve seen a spate of news stories and editorials based on a recent paper by David Bicchetti and Nicolas Maystre of the United Nations Conference on Trade and Development, “The synchronized and long-lasting structural change on commodity markets: evidence from high frequency data” (March 20, 2012).
The basic observation of the paper is that correlations between stocks and commodities, especially oil, rose in the second half of 2008 and for the most part have stayed high since. This is true, and well known. It is the conclusion the paper draws from the observation that is newsworthy:
“This result is important for at least two reasons. First, it questions the diversification strategy and portfolio allocation in commodities pursued by financial investors. Second, it shows that, as commodity markets become financialized, they are more prone to external destabilizing effects. In addition, their tendency to deviate from their fundamentals exposed them to sudden and sharp corrections.”
How does a 2008 regime shift in correlations support such strong conclusions about portfolio strategy and market dynamics? Of course the answer is that it doesn’t, but it’s instructive to see where the empirical data leave off and the opinions get inserted.
Correlation is a measure of how two price series move together. For example, the S&P 500 and the Nasdaq index have a 0.88 correlation over the last 10 years.
A correlation of one would mean the series move up and down in perfect synchrony. Knowing what the S&P 500 did on a day would allow you to predict perfectly what the Nasdaq did. So the S&P 500 and Nasdaq tend to move up and down together for the most part, but they are not perfectly in synch.
A correlation of zero means there is no linear relation between the series. A correlation of negative one means perfect synchrony but in opposite directions, when one series goes up, the other goes down. Since September 2008, the S&P 500 has had a 0.54 correlation with crude oil prices, while in the 10 previous years the correlation was zero. The chart below shows the annual correlations going back to 1970:
You can see that most years, the correlation is near zero. The correlation was negative during the oil shock from 1973 to 1976. This is what you expect from supply side volatility. If oil prices go up because supply is restricted, the higher prices hurt the economy, and the S&P 500 falls.
This is true even more sharply in 1990 during Gulf War I because in addition to the supply side argument, higher oil prices would likely have been the result of bad military and political news for the developed world (military failure in the Gulf War, or political failure to stabilize the region). The other significant negative correlation was Gulf War II.
The positive correlation in the late 1970s and early 1980s logically could have resulted from demand side volatility. If oil prices go up because the economy is growing fast, oil and the S&P 500 should go up and down together.
You do see this phenomenon before 1970, but since then economic growth has generally not meant increased demand for commodities.
Progress has been driven by technology, substituting knowledge for physical things. There are far more telephones with far more capabilities today than in 1970, but the total weight of all telephones today is less than in 1970, and the amount of commodities used to construct them is far less. Some people have argued that the industrialization of China and other emerging markets will restore the pre-1970 situation, economic growth based on increased consumption of commodities rather than technological progress.
I don’t agree with this, but it is one explanation for the recent increase in correlation between stocks and commodities.
The reason for positive correlations in the late 1970s and early 1980s, and in 2001 and, I believe, in 2008 to the present, has nothing to do with either the stock market or commodities, and everything to do with the dollar. The prices of both assets are measured in dollars.
If uncertainty about the value of the dollar is a major driver of market volatility, then stocks and commodities and everything else measured in dollars will move up and down together. Back in 1980, the uncertainty came from high and volatile inflation, which threatened to spiral out of control.
After the bursting of the dot-com bubble and to a much greater extent after the failure of Lehman Brothers, uncertainty came from excessive private leverage, government budget problems, fears of government shutdown or default, low interest rates, and increased government interference in the economy.
Bicchetti and Maystre focus on what they call “high frequency” correlations, measured over periods of 15 seconds to five hours.
One immediate problem is that most people use “high frequency” to mean millisecond trading. Fifteen years ago you could get away with calling trading every second high frequency, but things have changed. So the news reports claiming that high frequency trading is destabilizing commodity markets are really talking about all intraday trading, not just the people with co-located servers.
Although they have data going back to 1997, almost all of it comes from 2008 and later. Therefore, we are comparing sparse historical data, mainly from 2007, to rich data from 2008 to 2011. The problem is particularly severe for the highest frequency data, which shows the strongest effects.
You have to be careful when you see a strong effect in your data at the same time that the type or amount of data changes. You may mistake a change in your data with a change in reality.
In any event, the high frequency correlations confirm what everyone else has observed using daily data: that the correlations of stocks with other financial assets, especially oil but including other commodities and even the euro, went up in October 2008, stayed elevated for nearly two years, than began to decline but are still much higher than historical averages. The decline has not been smooth.
Most other analysts discuss this in the context of other historical regime shifts in asset return correlations, but Bicchetti and Maystre don’t have much data before 2008, and none before 1997, so they are unable to do this.
So how do they get from the observation that asset return correlations increased in 2008 to the conclusion that high frequency traders are destabilizing markets?
They begin by considering and rejecting four alternative explanations. First is the China argument described above, that economic growth in emerging markets will be accompanied by increased demand for commodities, linking growth and commodity prices globally. Under this explanation, the change in 2008 was that growth from emerging markets suddenly seemed to make up a larger part of world growth prospects than had been previously thought.
While I do not accept this argument, Bicchetti and Maystre reject it only because it “does not provide any satisfying explanation for the lasting co-movements observed afterwards.” Huh? If you believe ending poverty in the world requires consuming more commodities, why would your belief have changed since 2008?
The second explanation considered is that investors buy all risky assets when they think things feel safe, and sell all risky assets when they get worried.
Therefore, all risky asset prices go up and down together. I think this is clearly part of the explanation, although I would phrase it as investors demand different premiums for risk as uncertainty about the future waxes and wanes. The change in October 2008 was a massive increase in the premium demanded to hold risky assets. The paper mentions this explanation, does not offer any refutation, but then forgets about it.
Third, the authors consider inflation fears, the main driver of the positive correlations from the late 1970s and early 1980s. Again, they note the explanation but offer no counter-argument.
The fourth explanation is that massive market intervention by governments and central banks drove investors in and out of all markets for risky assets together. The rebuttal for this is the same as for the first explanation: that this does not explain why correlations remain high since the economy began to recover in the second quarter of 2009.
But the alleged economic recovery did not slow the massive tide of regulation or proposed regulation (including transaction taxes, rules against shorting and increased tax rates), nor the aggressive policy actions of central banks.
One obvious explanation, which the authors do not consider, is that increased trading activity allowed them to measure correlations which had been there all along. With light trading, prices in a market get stale, which suppresses measured correlations.
The problem is larger the more often you sample. That’s why professionals typically adjust correlations. With greater trading volume you would expect higher measured correlations, with the biggest effects at the highest frequencies. This is exactly what we observe. I suspect this contributed to the observed effect.
The real reason for the paper’s conclusions is what it says after going through the four explanations:
“Indeed, the recent price movements of commodities are hardly justified on the basis of changes of their own supply and demand. In fact, the strong correlations between different commodities and the S&P 500 at very high frequency are really unlikely to reflect economic fundamentals since these indicators do not vary at such speed.
Moreover, given the large selection of commodities we analyze, we would expect to have different behaviors due to their seasonality, fundamentals and specific physical market dynamics. Yet, we do not observe these differences at any frequency.
In addition, the fact that these correlations at high frequencies started during the financial shocks provides additional support for financial-based factors behind this structural change. Therefore, the very existence of cross-market correlations at high frequencies favors the presence of automated tradingstrategies operated by robots on multiple assets.”
This is an ancient prejudice common among macroeconomists. Real news comes out quarterly, with earnings releases and government statistics. Macroeconomists can compute the proper prices of assets based on “supply and demand” and other “economic fundamentals” that don’t change so fast that your analysis is worthless before it’s published.
Moreover, computing the proper price requires factoring in “seasonality, fundamentals and specific physical market dynamics” so if different types of assets move up and down together they cannot be doing the computation correctly. In short, market volatility and correlations at short time scales are bad because they’re not how macroeconomists think the world should work.
Another prejudice, more common among bureaucrats than economists (the authors are both), is that if something is bad, someone or something outside the organization must be blamed. In this case, the financial system is convicted because the change occurred at the same time as financial shocks.
Note that the paper does not blame the shocks themselves, which might suggest inquiring further about the cause of the shocks. The logic is simply anything that happened in the fall of 2008 can be blamed on markets.
This argument, technically called post hoc, ergo propter hoc (after this, therefore because of this), is a classical fallacy. It’s particularly unconvincing in this case because lots of things happened during that tumultuous period, including events related to the other plausible explanations for the correlation change.
There was a presidential election in the United States, a massive bailout, and chaotic political and regulatory events throughout the world. And there have been plenty of other financial crises that were not followed by increases in asset return correlations.
Everything comes down to the very last sentence. “The very existence of cross-market correlations” suggests the culprit is robots with trading accounts. No evidence is necessary, where there’s smoke there’s fire, and where there are cross market correlations there are robots. It’s not clear why we waded through the rest of the paper if it was all going to reduce to this.
Consider the plausibility of this explanation compared to the ones the authors reject. After the failure of Lehman Brothers, one or more robots decided to start cross-market momentum trading (that is, buying oil when stocks ticked up, and buying stocks when oil ticked up).
They decided to do this in all markets at once, no ramp-up. No one had thought of this strategy before, and no extra people joined in afterward. Two other explanations were rejected because they only explained the appearance of the cross market correlations, not their constancy over time.
This explanation explains neither one. There’s no reason for people to have started doing this in October 2008, and if they were successful, they should have done more and been imitated; if they were unsuccessful, they should have stopped.
Suppose we accept this explanation, weak as it is. How does it lead to the charge that high frequency trading has destabilized markets and driven them away from fundamentals? It doesn’t.
That is pure assertion. How does it lead to the other main conclusion, that investing in commodities does not provide diversification for equity investors? It’s true that higher correlations among asset returns reduce the value of diversification. But diversification still helps quite a bit for correlations on the order the authors observe.
More important, investors care about longer-term correlations, not one-second or one-minute values. The paper presents no evidence on those, so it has no relevance for the discussion. In fact, longer-term asset return correlations have increased, but not suddenly in fall of 2008, and not out of line with historical ups and downs.
After all that, we’re left with an inflammatory press release based on pure prejudice, appended to a straightforward analysis of data that confirms a well-known effect. Perhaps someday business reporters will read papers before writing their stories about them.
Read more: Has High-Frequency Trading Destabilized Markets? | Commodities And Options | Minyanville.com