Correlations with the S&P500 might now be 0.86, while ten years ago they were 0.14," says Jeffrey Rubin, research director at Birinyi Associates, "So there's been a shift, the question is why. The increasing popularity of indexing strategies is part of it."
Stuart Theakston, head of research and automated trading at GLC, stresses that we need to distinguish the short-term effects, where correlations rise when market sentiments decline, from a long-term shift in investment fundamentals. "ETFs have changed the market structure," says Theakston, "which is bad news for long-only asset managers, at least in the short term, if trading is based more on what is happening to other assets. The market won't recognize your views of value."
In their recent analysis of the current crisis1, Harald Malmgren, CEO of Malmgren Global, and Mark Stys, chief investment officer of Bluemont Capital, argued that at a minimum, computerized high-frequency and algorithmic trading were undermining traditional value investing strategies. "In effect, individual traders are confronted with overwhelming momentum-driven forces that are unrelated to performance of individual businesses.
A 'fair price' may exist, but high-frequency traders are not seeking fair prices - they are focused solely on immediate profit."
"A 'fair price' may exist, but high-frequency traders are not seeking fair prices - they are focused solely on immediate profit."
They argued that while high-frequency trading had an amplifying effect on market moves, it would be simplistic to consider it as the 'sole cause' of upward bias. In particular they emphasized contextual factors like pessimistic economic sentiment, deleveraging, declining 'natural' flows of liquidity from pension plans and a much smaller number of listed companies for investment.
"We have the perfect storm of macro economic pessimism, high levels of investment indexation and increasing automation using similar cross-asset strategies," says Theakston, "so correlation rises, but eventually this will pass over."
Sentiment and Volatility
The macro-economic outlook is clearly a key factor. "When people are driven by sentiment, you get pyramid schemes," says Rubin at Birinyi. "First gold goes up as an inflation hedge. Then gold rises as a deflation hedge. In the end gold rises because gold rises, until of course it falls, and sooner or later it will."
"Then when the economy is dire we see 'risk off' behaviours," adds Theakston, "whereas when sentiment improves there's more of a 'risk on' attitude." [Ed - See Marc H. Malek's description of 'risk seeking' and 'risk averse' environments in Me and My Machine from the Q4 2010 issue of Automated Trader Magazine]
In a research note of 26 July 20102, Maneesh Deshpande and Rohit Bhatia of Barclays Capital explored the relationship between equity correlation and volatility. They concluded: "While much of the variation in equity correlation is driven by equity volatility, from a long-term perspective it appears to have had a secular increase in value." They showed how the excess volume invested in index funds and ETFs compared to the value invested in their underlying constituent stocks rose steadily to 2005 but since then has traded "in tandem with equity correlations".
They drew parallels between option-implied correlation and realized correlation, similar to the dynamics of option-implied volatility and realized volatility, reviewing the evolution since the 1950s. After the bursting of the tech-bubble and during the subsequent telecom recession, Deshpande and Bhatia described how correlation at first rose but then fell again during the 2003-2007 rally. In particular, they stressed that while equity volatility dropped to the same levels as during the early 1990's, correlation did not and appeared to have been reset to a new level. "in our opinion this is driven by the increasing popularity of passive index indexing and (broad market) ETFs," they argued.
"Option-implied correlation," they explained, "which has traditionally traded at a premium to realized, is currently trading at a discount which would indicate that the option market is loathe to believe that the current high correlation is likely to persist. However, the fact that long-dated implied correlation is also equally high and almost equal to its short-dated version appears to provide a conflicting signal."
The Barclays Capital team resolved this conundrum by concluding, "while equity correlation continues to be highly dependent on volatility, the rise in indexation has led to a permanent increase in its 'base' level. Thus while we do believe that the current high levels of realized and implied correlations are unsustainable, the eventual drop is not likely to be as high as some market participants might expect." Theakston at GLC sees it very much as a cycle. "2003 to 2007 was a period of low volatility but also lower correlation," says Theakston. "Since the fall of Lehman, volatility has risen with risks, but correlations have increased as well. When we return to normality, fundamentals should reassert themselves."
The increasing use of quantitative benchmarks and index-based products has increased the dangers of what Professor Gautam Mitra, director of OptiRisk Systems, has called3 quantcentration effects. These arise where asset managers have access to the same sources of data or apply similar models.
"Back in 1974 there were few big events, like non-farm payrolls, in the trading calendar," says Rubin at Birinyi. "Today in any month there are over 100 macro data releases. Every day there's fresh news and everyone is comparing them to analyst forecasts. Inevitably there will be more correlation." He notes that over the past ten years there has been a huge up-take in ETFs, but the market is not going anywhere. "People became disappointed in active management, as most hedge funds don't deliver," says Rubin. "You have a thousand ETFs, some with three times leverage, others three times short, so prices change not because of fundamentals, but simply because money is shifting from one to the other."
"If more assets are pouring into ETFs than in the past," adds Justin Schack, director of market structure analysis at Rosenblatt Securities, "that has the potential to increase correlation, because every time an ETF is bought or sold and the price changes, it triggers moves in the underlying securities."
Rubin goes further, asserting that leveraged ETFs cannot work. "I'm shocked that regulators allow even 2:1 margin in the brokerage accounts with 3:1 leverage in the ETF," he says. "A small percentage swing wipes out most of your stake. That's gambling, not investing."
A swarm of models can have huge momentum.
"Back in the summer of 2007 market neutral algos fell apart," says Rubin. "Historically they might have been down 0.25% in a month. Suddenly they fell 6 to 7% in a day. There was a lot of money chasing a few shares. Everyone was standing on one side of the boat. Inevitably it capsized." He argues we may have 5000 hedge funds and many are still leveraged, but there are not 5000 different strategies. "Everyone is zigging and zagging together," says Rubin. "You can't have everyone outperforming an average. That's nonsense."
Now correlations are starting to go global. "People are now focused on continuous cross-asset hedging 24 hours a day, instead of hedging perhaps once a day on selected positions with large thresholds," says Nick Nielsen, head of trading at Marshall Wace. "While in the past they might have used CME's Nikkei futures in dollars, now they'll use a range of Asian instruments as well, including FX futures or commodities. The old world was very siloed, while now hedge funds trade anything. Such instant cross asset connectivity inevitably increases correlations."
High Frequency Hijinks
Malmgren and Stys see the positive feedback loops created by high frequency algorithmic trading as particularly destabilizing. "For decades, professional investment advisers have continued to teach reliance on 'value investing' and 'buy-and-hold' as long-term guides to successful investment. … Technology may now have overridden such investment concepts," they wrote. "In effect, individual traders are confronted with over-whelming momentum-driven forces that are unrelated to performance of individual businesses. A 'fair price' may exist, but high-frequency traders are not seeking fair prices - they are focused solely on immediate profit."
Rubin at Birinyi is similarly worried about the algorithmic traders. "We have too many traders leaving their programs in charge and going to the movies," he says. "They don't care about what they're trading as long as they go home flat. Inevitably it will fail."
Others however take a less dramatic view of technology. "The way an algorithm slices a block into pieces for execution over a few minutes, hours or even days can make a difference in how much it costs an investor to implement an idea," says Schack at Rosenblatt, "but it shouldn't affect the long-term pricing of securities. Suggesting that it does is a pretty big leap."
Schack argues that research shows there has been a dramatic decline in institutional transaction costs that coincided with the period of growing high frequency trading and algorithmic influence in the market. Much of the evidence suggests that automated market-making and arbitrage strategies make the market more efficient and reduce volatility. Of course, he agrees there are other factors that influence volatility, like macroeconomic conditions, and those caused spikes in volatility during the crisis.
"Reg NMS, decimalisation and other market reforms distributed liquidity across more venues and increased competition among exchanges and liquidity providers," says Schack. "With hundreds of firms competing to quote at the inside in price-time priority markets, speed has become more important. You need to be at or near the front of the line to get the trade. So now in the top 100 or 200 most actively traded stocks, there's barely a penny spread. That's a huge improvement." Of course he agrees there are pockets of illiquid stocks that will not see these benefits, but believes on the whole outcomes are better.
"During the depths of the crisis we saw massive selling pressure, huge investor redemptions, and prime brokers forcing deleveraging, with scarcely a glitch," says Schack. "That all testifies to the resilience of the markets, and even on May 6, the markets quickly recovered. Compare that to the fixed income or OTC markets where some shut down completely."
Perhaps it is less the speed of the market, than its homogeneity. "Money always flows to the successful strategy," says Theakston at GLC, "but when trades become crowded, another strategy becomes successful." He explains how automated strategies, both higher and lower frequency plays, could be having a short term effect as well. Machine rebalancing of portfolios and asset allocation, will tend to pick similar instruments or preference similar strategies. "Both fashion and regulation - anything that increases investor conformity - indirectly increases correlation," says Theakston. "In 2007 a lot of hedge funds were doing factor arbitrage. So they were overweight in mid cap, value stocks and underweight in growth stocks. Initially as more people adopt a strategy, returns improve. This carries on until the music stops, and everyone needs to unwind their portfolios in a hurry, and then everybody blows up at the same time. If everyone's using similar VaR methodologies, this accentuates it. This was less a matter of correlation and more getting into a crowded strategy that blew up completely."
Escaping the Loop
With all these forces of market sentiment, indexation, and quantcentration pulling us into the black hole of ever-higher correlation, how do we escape?
"Professional traders used to have a strong human element. Now it's more 'clerkified', says Rubin at Birinyi. "We have to break open the black boxes and start to think for ourselves. What can we do that's different? What's working and what's not?" Rubin also thinks we need to bring back the fiduciary duty of the market maker. "Anyone who sells an ETF should make a market in it and find others to commit liquidity as well," he says. "It won't stop all crashes -- there were plenty of specialists in the '87 crash -- but it should slow down the next one."
Rubin sees the May 6 flash crash as just a continuation of the last 10 years. "Quote stuffing, market makers pulling liquidity, the ease with which people can cancel an order - it's all too unstable," he says. "We need to increase the level of commitment in the market."
Similarly Malmgren and Stys want to engineer more balance into the system. "Given the fundamentally different objectives of high-frequency traders, their unlimited freedom of action, and their apparent dependence on positive feedback loops, regulators need to think beyond circuit breakers to devise compensating negative feedback mechanisms."
"However," warns Nielsen at Marshall Wace, "we need to be careful. The increasing correlations could be regime specific and linked to current uncertainty. In five years' time it might be much less macro driven."
So, perhaps all we need to do is weather the storm. "The competitive dynamic will of course eventually unwind," says Theakston. "Macro sentiments will improve, high frequency trading is itself arbitraging away its own advantage, and fundamentals will reassert themselves, even if ETFs are here to stay. Each effect is at a different stage of the cycle, but in the end it's just a cycle."
Theakston argues that more passive investors increase short run correlations, but should make it easier for other investors to be active. The contrarian view should increasingly differentiate itself. Similarly, as more investors become active, then a passive strategy becomes easier. "When we look only at recent history they seem to be coming together," says Theakston, "but a longer alpha horizon might suggest a different conclusion."
Theakston believes it important to have a sense of what others are doing, to see the big picture. "That's easier to say, of course, than do," he admits, "since people are inevitably not keen to talk until after the event. Indeed when they start to talk that signals the good days are probably over."
So perhaps since everyone is now talking about correlation, we should all be going back to value investing! If only the conundrum were that simple!