The best argument for a pullback on Wall Street may be the simplest: There hasn’t been one in a while.
With Monday’s modest declines, US equities were a little more than 2% from local highs. At the least, stocks are due to fall another 1% to 3%, according to history.
“The S&P sells off by at least 3-5% on average every two to three months,” Deutsche Bank’s Parag Thatte said, in a positioning and flows update. The last “meaningful” selloff was in March, around the mini-banking crisis, which means the current rally ranks in the 85%ile, as illustrated in the figure below.
We’re “arguably overdue,” Thatte remarked.
For readers exhausted with belabored attempts to justify a bearish narrative by way of tedious, if compelling, fundamental analysis and/or arguments centered around the idea that the marginal buyer is becoming harder to identify, “we’re due” is probably just as satisfying as any other rationale.
But the key point from Thatte’s update was the idea that the confluence of factors which conspired to pull in both systematic cohorts and, belatedly, discretionary investors, is probably out of gas in terms of its capacity to drive additional re-leveraging.
Lower index vol was driven in part by collapsing correlations, and lower portfolio vol is the key toggle for systematic investors. But correlations are now near record lows. “Only in late 2017- early 2018, right before the ‘Volmageddon’ selloff” were correlations lower, Thatte observed.
Dispersion associated with outlandish tech outperformance (and subsequent momentum reversals) was partly responsible, but correlation also tends to fall when macro considerations lose their influence on the market. Either way, it’s hard to see correlation collapsing much further. Any increase would be conducive to higher index vol, which in turn could have implications for systematics.
On Deutsche’s metrics, equity exposure for vol control, CTAs and risk parity sits in the 68%ile, 81%ile and 77%ile, respectively. The bank’s aggregate systematic metric is shown on the right, below. It’s in the 84%ile.
“For the dynamic of falling portfolio vol and rising leverage to continue, bond vol needs to break below the elevated range it has been in since last February,” Thatte wrote. While there’s a case to be made that short-end rates vol is indeed poised to keep receding, it’s also true that central banks across the developed world are in the process of extending their tightening campaigns. That could be conducive to additional policy uncertainty.
As for discretionary positioning, it’s not stretched in an absolute sense, but after catching up to systematics in recent weeks, it’s “elevated relative to its historical drivers, namely growth indicators like the ISM and PMIs, and has already priced in a soft landing,” Thatte went on, referencing the visual on the left above.
The implication is clear enough. You don’t need to go to any extraordinary analytical lengths to make the case for a selloff. It can be as straightforward (or not) as you want it to be.
I have all SP500 and NDX data from their inception and often perform analysis like these. The problem that I always run up against is what data to include. All data? or only data that are from periods that I “think” are similar to the current one? there will be fewer 2-3% pullbacks when market is moving about 0.5% per day than when it is moving 1% per day. Actually that is just a guess because I haven’t done that analysis.
what I mainly use these types of analysis for is to get me out of FOMO mode. Now is probably not the best time to be going all in on the longs or the shorts. I am much more in profit taking mode from positions from Oct-March. The worst feeling for me is getting drawn into FOMO and then getting immediately slapped down. Somehow I feel better and have better results if the SP moves up 3% from here pulls back 2% and my indicators say to buy, even though I am buying at a higher price. Everyone is different in this regard.
One way to answer your question is to select different data schemes and try them in your model(s). If there is a significant difference in the outcomes of alternative configurations, then the changes you are making are, in fact, critical and must be understood before reaching any conclusions. If the outcome is robust and leaves the problem with outcomes that are not significantly different, then your question is not interesting. I would bet that what data you choose in the related problems you are studying is definitely critical. The more interesting question to me occurs when one is assessing the research of others. As an editorial reviewer I do this regularly. When critical variables are present and a researcher can make changes at will, then there is a definite possibility that a researcher can get any income s/he prefers, especially when the reader does not know what data points were used, what was omitted, and what alternative models were tested. This whole question will be central to growing the usefulness of AI and to protect its reliability.