Warning: extreme geekiness ahead…
In my previous post I looked at the Russian stock market and showed that like India, but unlike the US, short-term price changes in Russia are momentum, not mean-reversion driven. That’s important for at least two reasons: (a) short-term indicators like RSI(2) will work opposite they way they do in the U.S., and (b) it completely changes the dynamic of whether we should let winners run or take profits quickly, and cut losses early or wait for trades to develop.
This post will be about two oddities in the Russian data that I know are important to the US market, but I’m still wrapping my head around why.
[logarithmically-scaled]
In my previous post, I showed a graph like the one above that assumed a trader went long (at the close) following a close up, and short following a close down, Russia’s RTS Index (denominated in USD), from 1996 frictionless (these results wouldn’t actually be achievable, but just for giggles, that’s a 132% annualized return). Note that based on these results, the Russian market is momentum-driven: up days tend to follow up days, and vice-versa.
ODDITY #1: BEND IN THE EQUITY CURVE
Note the bend in the equity curve in the graph above (denoted by red arrow). Right about that time, follow-through in Russia lost a good deal of its steam. It still existed, but was far less potent than the years prior.
That change occurred right around the turn of the millennium, which is of course interesting, because it’s right about the same time that follow-through in the US flipped from being momentum to mean-reversion driven (read more here and here).
So the question is: (a) was this roughly simultaneous change in Russia because of the influence of the US market? Or (b) did both change because of some third outside influence (more on this below)? Or (c) is this purely coincidence?
It would be interesting to do a study of how each market directly influences the other like I did with East Asia (read more). If there is a strong connection with the US (like Hong Kong), then a, b, or c above might be true. If there isn’t a strong connection (like in Shanghai), then only b or c might be true. More to follow on this.
ODDITY #2: INDEX DENOMINATED IN RUBLES
This is the one that really has my goat.
Compare the graph above of follow-through in the Russian index denominated in USD, to the one below of the same strategy applied to the index denominated in Russian rubles…
[logarithmically-scaled]
Note the breakdown in follow-through over the last year or so (red arrow). It’s a little hard to put in perspective because this is a logarithmic-chart, so below I’ve included a graph of the drawdown of this same strategy (i.e. how far the strategy is from its highs at any given moment).
Now the breakdown is a bit easier to see. Since about October of last year, follow-through in Russia has flipped contrarian (and is now like the US).
Is this change permanent? I don’t know, but if it is, it raises a very interesting question:
Do fluctuations in the strength/weakness of a nation’s currency have an impact on short-term momentum vs mean-reversion in the country’s stock market? Remember that the only difference between the first and second Russian tests were that the first was denominated in USD, and the second in Rubles.
We’ve seen the Ruble growing stronger over the last 6 months and a breakdown in short-term momentum in the last 9 – are these connected? We saw stabilization and even appreciation in the Ruble between 2000 and mid-2008, and over the same period, a tempering of short-term momentum – are these connected? Or are these both being driven by some third outside influence?
I don’t know the answer to any of these questions, but it’s uncanny how well the events line up.
I’m going to keep plugging away at this concept. Short-term US price movement is so important to what we do in our own proprietary strategies that anything we can do to model and better understand its ebb and flow (and when it will make its next evolution) is incredibly important.
Happy Trading,
ms
To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.
Filed under: Evolving Markets, Follow-Through, International Markets | 13 Comments
one theory is based on observations from Mark Weinstein one of the traders featured in Market Wizards. He first notes that the futures markets exhibit stronger trend persistence in the short term than the stock market–having traded both on the floor for a living. He claims that the reason for choppy price action in the stock market is that insititutions and specialists scale in and out of their positions over time—-leading to choppy price action. They wait for down days to buy and up days to sell. In contrast in the futures markets due to the extreme leverage and the contract sizes, and the fact that most players are system driven, players exit their positions in one shot—especially because of the possibility of being locked limit up or down. The choppy price action in the stock market (read mean reversion) if indeed is being caused by institutions is probably getting stronger because of the accelerating market share of insitutional trading vs retail trading. Bear in mind that before the 90s started, retail investors trading from independent brokerage accounts were a big part of the market. In fact prior to 1960, they were the dominant players in the stock market. Mutual funds and hedge funds especially gained most of their popularity in the 90’s. Hedge funds gained most of their popularity just at the turn of the millenium, and being sophisticated players, they tended to focus on pair/arbitrage strategies that have the effect of increasing mean reversion in the broader markets because they would short pepsi if it went up too much relative to coke etc in the short term. With the unlimited leverage given to them for free via Goldman prime brokerage etc and spectacular growth in assets from pensions etc……it is easy to see how this could be excacerbated. In fact if we look at futures markets following the introduction of the ability of pension funds to invest in commodity ETFs (USO, UNG etc) you will find that mean reversion has picked up in futures now as well—likely due to the same factors (scaled increasing or decreasing money flow). Interestingly small cap stocks do not exhibit such pronounced mean reversion, and until recently the Russell 2000 was a highly trending index. And who trades in small stocks? according to research it is primarily the retail investor. The small the market cap the better trend following strategies perform, and the more likely retail investors dominate the market. Developed markets like Russia or China etc, are chock full of retail investors—–this explains daily follow through—individuals pile into investments likely lemmings all in one shot with new ones buying at higher prices above them. They don’t have the size to be able to scale in and out nor the neccessity. As developed markets mature, asset management firms spring up and offer mutual funds and hedge funds etc and eventually as institutions become a big part of the market, mean reversion gets introduced.
this is just a theory
dv
Very interesting Michael. Nice work. I’m looking forward to seeing what you are able to determine as a possible cause, primarily as it might help us decode when the US markets start moving back towards momo.
Hmmm….makes me want to look at the dollar. The dollar started a strong decline around the same time the market became mean-reverting. Probably just coincidence, but I don’t have enough dollar data to look back to see if there is any correlation.
I used to run prediction market for a company – we saw very different behavior in the market when noise traders came in. Maybe it is a reflection of market maturity – they all start out, after opening, rather like an IPO – all up! Then the noise traders come in.
david, that might me just a theory, but a very interesting theory, indeed. I enjoyed reading it, thanks for the writing.
The first oddity might be explanined by the first Russian Financial crisis in 1998.
I’ve done a long ago a similar study for the Brazilian stock market (our currency was messed up before 1998) and we got pretty heavy structural changes since the Asian/Russian crisis…
Excellent post, David. Very thought-provoking. Thanks.
(and thanks to Michael for the original post asking the questions)
David, it seems intuitive (to me) that stat arb/pairs would reduce mean reversion as it forces stocks to trade very near their mean, never venturing to far above or below. In fact, look at the US indices between 2005-2007. It also seems to reduce volatility, which would hurt the performance of mean-reversion systems.
woodshedder you are correct, when i spoke of mean reversion with regard to funds my wording was intended to suggest exactly what you are describing which is a containment near the mean……i misused the term in the context of what we were talking about, the great deleveraging of the financial markets has greatly impeded funds ability to do pairs trading etc, which is exactly why ETF arbitrage spreads skyrocket as a function of volatility and the vix (i have excecuted several ETF arb trades–but few decent ones have been available in the last 3 months) . In fact the performance of basic stock pair strategies in the US had its best year ever last year, refecting the deleveraging. This is why daily follow through mean reversion has accelerated in both stocks and indices in 2008/2009. During the reliquification over the last few months, and the fact that Goldman raised capital (the biggest prop trading desk), probably explains the return to more normal volatility and reduced mean reversion. But the scaling effect via mutual funds is a major driver. With respect to developed markets Brazil etc, pair trading strategies have much higher sharpe ratios over the last 20 years reflecting the lack of institutions engaging is such strategies. However over the last 10 years, sharpe ratios have declined signficantly in developed markets—perhaps because of domestic hedge funds or north american funds expading into their territory.
To determine the success of mean reversion strategies in the future, a basic pair trading strategy performance index —using either Dow 30 or S&P500 stocks could be constructed using RSI2 differentials. If the performance is above average or accelerating, that could predict future increased mean reversion, the reverse would be true in the opposite case.
Similarly an index of aggregate assets in domestic equity mutual funds/vs retail traders using a baseline could also predict the same.
dv
i forgot to mention the the other bread and butter of prop trading desks is index arbitrage relative to futures which requires extreme leverage–this causes the index itself and its components to exhibit similar behavior. Also the disapearance of NYSE specialists hasn’t helped either, as they tended to aggressively fade moves intraday, thus muting moves in major stocks. day trading strategies also had their best year last year as a result.
Hi,
I wanted to know how to compute how similar two systems are. I have developed 3 systems based on daily timeframe. They are all at any time either long or short. The first and second are very similar – in the sense its just the tweaking of one parameter which i believe leads to sufficient diffrentiation to call is a diff system, or treat it as one. The third one is completely different. I want to know how to determine how different they are, or what is their correlation to each other (or such parameter of all performing badly together). I dont know if correlation between the long and short part is a good measure, or whether, a test such as how many % of total no of trading days were they in the same direction is a better measure. I tried the second measure, it gave me values of 66% between the first two systems ( the similar ones), and 75 & 76% between 1&3 and 2&3 respectively. So how do i interpret this ? Good enough to implement all 3 simultaneously ? Especially since the similar systems have a smaller % of same direction than with the 3rd one. And the % for all 3 in same direction was just 25%.
Can you write a post on this, or simply reply back with a comment.
Thanks :)
RE to piyush: very good question…am I correct in saying that we’re trying to figure out “how similar they are” and not “% to invest in each one”?…if so, some thoughts…
You hit on some of the low-hanging fruit: daily/monthly correlation and % of days trading the same direction. I also like to look at correlation between programs X, Y, and Z only on days when program X loses money (rather than all days…sort of a “losing correlation”). Also correlation during different market themes (bull, sideways, and bearish markets).
The more difficult question of course is how to translate all of that into a % allocation between each program…I’m still mulling a lot of thoughts on this one in my own head.
michael
yes ..how similar they are.. il try to figure the second one out later..
and for daily correlation and % in same direction, and correlation on losing days – what figires do you think would qualify as good, average, poor etc.
thanks :)
RE to piyush: IMHO, I don’t think there’s a right answer. It’s like the traditional “efficient frontier” concept – you want as little similarity as possible while still employing successful individual strategies. michael