Charting The ETF And HFT-Derived Record Correlation Bubble
JPM's Delta One team has come up with some great observations on what is the one truly indisputable bubble in the market currently: that of correlations (unlike the bubbles in bonds and stocks where both camps have stern defenders who refuse to acknowledge that values are only where they are due to the Fed's now daily intervention). Global Head Marko Kolanovic also provides some interesting observations on how HFT is responsible for this record surge in correlations.
Correlations between stocks are currently at the highest level in recent history. This is a result of the macro-driven environment, record use of index derivatives such as futures and to a lesser extent ETFs, and high-frequency trading. The option-implied price of correlation is even higher – a result of an inadequate supply of index put options and oversupply of stock options via overwriting. We believe that correlation levels are in a bubble-like regime and are bound to decline.
On realized and implied correlations:
Correlation measures the degree to which prices of stocks move together. The average correlation between all S&P 500 stocks is currently at historically high levels. In particular, the level of correlation recorded over the past two years was never realized in the recent history of U.S. markets (Figure 1). Two recent bursts of correlation (following the Lehman default and during May 2010) were matched in intensity only during the market crash of 1987. While the levels of correlation ebb and flow with business and volatility cycles, the average level of correlation was gradually increasing over the past ten years, even prior to the 2008 credit crunch (Figure 1).
The pricing of index options relative to options on individual stocks implies the level of market correlation. Historically, correlation priced in the options market was higher than the actual realized correlation between the stocks due to excessive demand for index protection. The current level of premium of option-implied correlation (over market realized correlation) is also close to historical highs, as shown in Figure 2.
On the drivers of market realized correlation:
A significant driver of correlation between stocks is the prevailing macroeconomic environment. During periods of high macro uncertainty, stocks prices are largely driven by macro factors such as economic growth, unemployment, interest rate changes, inflation expectations, etc. Therefore, during changes in macroeconomic regimes, stock prices tend to move in unison leading to a high level of correlation. Periods of high macro uncertainty are also characterized by high equity volatility. Figure 3 below shows the regression of correlation of S&P 500 stocks, against market volatility. While the relationship between correlation and volatility is strong, occasionally these two measures have diverged. For instance, during the inflation and burst of the Technology bubble, stocks were quite volatile, yet correlation was low due to a strong divergence between stocks in the ‘New Economy’ (Dot-coms, Technology stocks) and stocks in the ‘Old Economy’ (e.g., Utilities, Industrials). This type of intersector performance divergence caused overall correlation to plummet relative to average stock volatility. Today, we appear to be in an opposite correlation regime: stocks exhibit the same level of volatility as during the Tech bubble, yet they are all driven by the macro outlook for the economy and hence exhibit extreme levels of correlation (Figure 3). The comparison of these two regimes motivates us to deem the current environment a ‘Correlation Bubble’. Figure 4 shows the excess correlation1 over market volatility, which can give us a better historical perspective of the run-up in correlation over the past ten years. While correlation has increased steadily over the past ~10 years (2000- 2010), the first ~5 years of increase essentially brought correlation from the Tech bubble lows to a historical average level. However, over the past five years, correlation has been increasing more rapidly than implied by the macro environment (market volatility), pointing to the existence of additional drivers of correlation.
On ETFs as drivers of increased correlation:
We believe that the excess levels of correlation are related to the increased usage of index-based products, in particular futures and some broad-index ETFs. When investors trade an S&P 500 futures contract, they effectively place a simultaneous order for the 500 constituent stocks (e.g., buying a future will cause incremental upward pressure for all 500 stocks, and selling a future will cause an incremental downward pressure for all 500 stocks). It is easy to see that if investors only traded futures (e.g., futures were 100% of all equity volumes) the correlation of stocks would be 100%. For this reason, it is reasonable to expect that market correlations should be proportional to the prevalence of index products relative to stock volumes. Broad-index ETFs (such as S&P 500 ETFs) have a similar effect on market correlation.
Figure 5 below shows futures and ETF volumes expressed as a percentage of total cash equity volumes. One can see that over the past ten years, trading of index products experienced significant growth relative to stock trading. In particular, the share of futures and ETFs steadily grew over the past five years, and is now ~140% of cash equity volume (i.e., futures and ETFs are roughly ~60% of all equity volumes – perhaps not a coincidence that realized stock correlation is ~60%). The growth in index volumes coincided with a rise in correlation over the past ten years. More importantly, the growth of index volumes is directly driving excess market correlation (levels of correlation above the levels implied by macro volatility). Figure 6 shows the excess level of market correlation and S&P 500 futures volume. We note that the excess market correlation closely follows the ebbs and flows in S&P 500 futures usage. We believe that futures have a much larger impact on the market correlation than ETFs. The main reason is that futures notional volumes are significantly higher than ETF volumes (futures volumes are approximately double ETF volumes) and not all ETFs lead to an increase in market correlation. Currently about 60% of ETF volumes are in broad-based index ETFs that do contribute to increased correlations. However, almost ~30% of ETF volume is in sector ETFs, or ETFs with a significant sector bias. While these ETFs may lead to an increase of intrasector correlations, they may lead to a decrease of correlation between sectors (intersector correlations) thus reducing the overall average correlation between the stocks. Finally, 10% of ETF volumes are in commodity or fixed income ETFs that will have little impact on equity correlations.
On HFT as drivers of increased correlation:
Over the past several years, program trading and in particular High-Frequency Trading (HFT) experienced strong growth. Figure 7 below shows NYSE program trading volume and Figure 8 estimated total HFT volume in the U.S. It is estimated that currently close to 60% of U.S. turnover by volume is due to HFT (in Europe, it is estimated that ~38% of trading volume is due to HFT). It is reasonable to expect that such magnitude of trading activity will significantly change the market microstructure. We believe that High-Frequency Trading activity has increased correlations, reduced volatility, and increased the intraday tail risk. In order to understand how HFT activity can impact the market, we will look at two common HFT strategies: index arbitrage and optimal execution of orders. Index arbitrage is an example of HFT arbitrage trading. As shown in Figure 5, current index volumes are significantly larger than total cash volumes, and a good amount of index derivative volume (Futures, ETFs) will not be directly offset by trades in cash securities. If the index price diverges from the prices of underlying constituents, index arbitrage HFT will act to realign them. For instance, if a group of stocks outperforms the index, an arbitrage program may sell these stocks and buy the index, causing their prices to realign. This trading activity will dampen the volatility of stocks and increase their correlation to the rest of the stocks in the index. HFT index arbitrage also facilitates the transfer of the market impact on futures and ETFs to the underlying stocks, thus providing a link between the high percentage of index trading and correlation of individual stocks. Another HFT trading strategy is a statistical arbitrage. A simplified example is a pair trade between two correlated stocks. If the price of one stock increases relative to the other, an arbitrage program will sell the outperforming stock and buy the underperforming one, thus reducing the volatility of both and increasing the correlation between the two.
HFT also provides liquidity by breaking up larger orders and optimally allocating smaller orders across multiple sources. This activity will generally reduce the market impact of large individual orders and hence reduce the amount of stockspecific volatility. Reduced stock-specific volatility will result in lower dispersion and hence increased levels of correlation.
HFT is frequently quoted as a cause of market volatility. This is not justified in our view. As we showed in examples above, a typical HFT strategy may dampen volatility and increase correlation, which is consistent with the current record spread of correlation over volatility. Perhaps a reason for connecting HFT and market volatility is because HFT strategies thrive during volatile markets, when the stock moves captured by HFT are significantly higher than the transaction cost to execute a high-frequency trade. However, we believe that HFT may increase the probability of an intraday tail event such as the flash crash of May 6th. HFT liquidity-providing strategies essentially follow a set of trading rules that were designed to work in a normal market environment. As a HFT strategy is executed by computer code, it can not interpret significant fundamental news such as results of a court ruling, a large order imbalance, or the likelihood of a sovereign debt crisis. To protect trading profits, HFT strategies are likely to implement circuit breakers and pull liquidity as a response to significant idiosyncratic outliers that can’t be interpreted by the HFT code. The result of a sudden pull in liquidity by many HFT programs could be a sharp drop in the price of a security (or the whole market) similar to what we saw during the flash crash.
In other words, record correlations, which usually precede an all out market collapse (which hasn't yet occurred just because the Fed still has some tenuous credibility that it will prevent the crash thus preventing mass buying from becoming mass selling) are there precisely due to the SEC's complete ignorance, and allowing ETFs and HFTs to become the primary components of market structure: as we have long claimed to be the case. And the longer we continue without doing anything about it, the greater correlation will become, until one day not even the Fed's backstop can prevent stocks from going from whereever they are to nothing in the blink of an eye. Then, and only then, perhaps, will investors realize that the SEC is the most corrupt, technologically backward, lazy, and overpriced government agency ever conceived, and that both ETFs and HFTs pose a by far greater systemic risk than some unnamed fund manager out of Kansas. But for that to happen the next flash crash, which however will not have an up-leg to it, must occur, and far more than just $1 trillion in market cap has to be wiped out. And it will occur, as correlations creep steadily toward one: it is not an open ended equation, as luckily there is a limit to the next and final collapse.