
kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


The Opportunity in Big Market Swings
WHEN THE BUBBLE BURST in 2000, it seemed that a very longterm cycle in the stock market was simply asserting itself. For the Dow Jones Industrial Average, a pattern became evident that suggested the first two decades of the new millennium would result in big swings both higher and lower. Indeed, we have seen one of each since 2000.
While a buyandhold strategy may not be the best plan, there should be huge opportunities for investors along the way, such as the current rally has been from the 20022003 lows. As long as investors know when it is time to move to the sidelines, and that could be soon, the stock market should be embraced, not feared.
Before continuing, we must differentiate between two types of markets. The first is called "secular" and its commonly accepted definition is "long term," with a time frame measured in years. The word itself is derived from the Latin word for "age" or "long period of time" so we can understand the spirit of a secular bull or bear market. Some call them "generational" markets.
The second type of market is called "cyclical" and its commonly accepted definition is "short term," with a time frame measured in months. Cyclical ups and downs occur within secular markets; to many, the bear market of 20002002 was a cyclical move. It follows that the rally from there has been called a cyclical bull market.
With this in mind, we can look at the market with both long and shortterm eyes and keep each in perspective.
Charles Dow, cofounder of Dow Jones & Co., parent of Barron's Online, first published the DJIA on May 26, 1896. When the entire history is plotted on a single graph, a cycle of approximately 18 years of rally followed by 18 years of consolidation comes to light.
For much of its early life, the Dow traded in a choppy range between 40 and 100, hitting the low end of its range for the last time in 1914. From there, it began the rally that ultimately culminated in 1929 at a peak price of 380.
Of course, the Crash of 1929 sent the market reeling and huge multiyear swings both higher and lower were the norm. These were cyclical bull and bear markets within a secular period consolidation.
Finally, as the market calmed down, it broke out from a triangle pattern in 1945. The secular rally then continued in an orderly, yet powerful, manner until peaking in 1966. That is 16 years of turbulence followed by 21 years of rally and well within acceptable variation for the cycle.
The next period of turbulence included the great bear market of 197374. The Dow remained stuck below the 1000 level for 17 years until finally breaking out in 1983. From there, most investors know that the next great secular bull market lasted 17 years to the end of the bubble in 2000.
Therefore, if the pattern holds, large ups and downs in the form of a giant consolidation pattern should be the norm until late in the next decade. But that should not scare investors. Even in the 1970s, when the stock market was fraught with problems and trapped within a secular range , there were three sizable cyclical rallies. This included a twoyear bull market in 197576 that just about erased the bear market of 197374.
In recent weeks, Getting Technical has presented evidence that the current cyclical bull market is not only mature by historical standards but is starting to exhibit similarities to market peaks or near peaks in 1994, 1999 and 2004. While the 18year cycle does not pinpoint when the current cyclical bull will end, it does suggest that there will be an important decline coming.
Cyclical bull markets are followed by cyclical bear markets.
The main uncertainty is what shape the current secular period will ultimately take. The 1930s were a period of calming down from a crash to normalcy. The 1970s were a more or less flat period with regular ups and downs. And so far, the higher peaks set by many indexes above their respective 2000 highwater marks could be part of something completely different. It could signal a steep decline to come or just a mild pullback lasting many months if not years.
But even though big declines can be expected in the coming years, investors can take heart in knowing that a cycle is at work Cycles mean rallies follow declines. To ride out the bad times and prosper in the good, a shift from a strategy of "buy and hold forever" to one that encompasses market timing on its most macro level can be the best plan for even conservative investors to follow. 
my bank got pwnd 



kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


I've done some more runs with the topten ETFs from the list  I think that means I left out USO, GLD and EEM for lack of extended history. 10 assets seems to be a decent starting point because allocationspace searches are going to be pretty computationintensive beyond this level. The UP algo does OK but I only ran it on monthly data. I don't know how the stabart people think about this  if you are rebalancing on a monthly basis and you are looking for a firstorder algorithmic strategy, how much am I missing by leaving out daily data?
Anyhow, on that basis we're looking at 4.5 years of data i.e. 50 data points. I must admit I was a bit disappointed with the UP runs  basically my weights were still largely equalweighted, and the overall performance was not hugely better than equalweighted. It goes back to what I said before  if you're looking for the best portfolio among constantweighted, then it may be that your benchmark is a bit crap. For rebalanced portfolios and real 'information content' you can do the following exercise: Assuming perfect foresight, suppose you chose the best performing asset or the worst, each period. This will give you the absolute best and worst performance over the period. What you will see is that the range is far wider than the best and worst nonrebal portfolios. In fact, if you managed NOT to choose the worst two performers each period, and just equalweighted the rest, you would do FAR better than the UP algo. You might think that that kind of hurdle might not be so difficult to achieve.
Anyhow the article below is from Barrons and indirectly says the same thing, especially in today's context where vols are picking up. So I'd like to make this thread into a bit of a challenge: Come up with the bestperforming algo and some underlying foundational reasoning, where the assets are my top10 ETFs, and rebalancing is done monthly. Any extra timeseries can be included as long as there is a reasonable relationship present that can be explained, the data can be gathered in a public way, and belongs to the right information set in terms of forwardlookingness. My performance benchmark is something like best performance vs. worst drawdown or some other relatively simple metric, though I'm steering away from Sharpe here. I will post some benchmark statistics for context on this dataset soon. 
my bank got pwnd 



KR, you asked "To be contrarian, can you choose the weights so that the portfolio vol remains nearly constant throughout?"
I'm not sure that this is contrarian  although not my universe, a friend of mine who used to run a fairly large portfolio had second benchmark (after absolute return) driven off the volatility of volatility of returns. Basically, the fund wanted the "risky" allocations to have constantly high volatility, and the "safe" allocations to have constantly low volatility. Rather than force investment styles on portfolio managers, they just benchmarked them.
One thing that we discussed was to use the options market to estimate your future portfolio volatility by decomposing the position into a portfolio of vanilla options, then using current market implied volatility surface and portfolio weighting to estimate the future volatility of returns.
Not sure if he ever got it done  must ask him next time we speak.
Anyway, as I said, not my universe, so I have no idea if this was a) a stupid idea  everybody does it already b) a stupider idea  somebody tried it once, and got carried out c) of some merit
EQ
edit : typo fixed 





One thing that we discussed was to use the options market to estimate your future portfolio volatility by decomposing the position into a portfolio of vanilla options, then using current market implied volatility surface and portfolio weighting to estimate the future volatility of returns.
The only problem I can see is predicting correlations. Implied vol would give the expectations of vol for a single stock. When you have a basket of stocks you would also need the correlations to estimate the vol of the basket. Or maybe I am missing somthing...




kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


I am gonna keep bumping this up b/c I want to make this work, and get some opinions from people on how to approach this. I've been delayed on my range stats b/c my computer had a bout of cardiac arrest this week, but now we're back in business. Some thoughts:
 That 'volatility pump'... i.e. the stock that oscillates between +50% and 50% or whatever it is. Ok, it IS true that through constant rebalancing, you can make constant profits, but let's look at the numbers: If you rebalance each year, then 'through the cycle'  i.e. two periods, you make ca. 4%  i.e. 2%/year. On the other hand, if you KNEW the stock was a pure oscillator, you'd put all your money in the riskless asset on the way down, and all your money in the risky asset on the way up. Outcome there is clearly +50% over the cycle, or 22%/year. My point is that constantproportional weighting is not exactly the most admirable subset of strategies to look at. This is especially the case if we are trying to stay within bank compliance holding periods of at least 1 month.
 It also leads me to ask a simple question on stab art: How much of the problem boils down to identifying whether the 'mode' is either a) trending or b) oscillating, and some sort of phase estimation? Would anybody say this is the 'first principle of stab art'?
 Taking that further, this is an analysis of two assets, where 'rotation' is just a binary. When there are multiple assets, the modes of oscillation could be much more interesting (especially over longer sampling times). For instance, with an oscillation period of say 6  i.e. assets doing sine waves which complete their cycle in 6 units  then the maximally dynamic strategy is pretty clear to see: Put your money on the asset with the greatest positive derivative when you can, or in the riskless when you can't. I suspect I could find a phase arrangement such that constantproportional weight might actually be a net loser.
 Finally, the information set, and volume: Say we are running an ETF strategy... then in fact volume is easily observed. I'm thinking that I need to enlarge my info set a bit, would people say volume is the second thing to look at?

my bank got pwnd 



schmitty


Total Posts: 57 
Joined: Jun 2006 


That 'volatility pump'... the stock that oscillates between +50% and 50%
No, it doesn't oscillate, it has an equal probability of +50% and 50%, there is a huge difference. And in Shannon's original formulaton, the + number is larger than the minus number.
If you rebalance each year, then 'through the cycle'  i.e. two periods, you make ca. 4%  i.e. 2%/year
No, There is no constant proportional bet that makes money at the +50/50 numbers on a pure oscillation. At +50/50 any constant proportional bet requires some runs just to have a breakeven expectaion.
identifying = trending or oscillating 'mode'... anybody say this is the 'first principle of stab art'?
No, not even close.
I suspect I could find a phase arrangement such that constantproportional weight might actually be a net loser.
Yeah, like ALL of them, under your +50/50 pure oscillation scenario.
would people say volume is the second thing to look at?
No.




kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


If I may quote from Blum/Kalai, 'As an example of a useful CRP, consider the following market with just two stocks... The price of one stock remains constant and the price of the other stock alternately halves and doubles... On alternate periods the total value will change etc...'
http://www.springerlink.com/content/m01l6704j8377w67/fulltext.pdf
The point of the Blum/Kalai thing was that the + figure didn't have to be bigger than the  figure. I take your point that it could be 50/50, but when it DOES oscillate (i.e. doesn't have runs) then rebalancing DOES make money. That's why I brought the whole thing up.
As for 'first principles' I'm skimming through the Park/Irwin 'Profitability of Tech Analysis Review' thing which seems to be a pretty decent overview. I think it's not totally unreasonable to draw a parallele of the Blum/Kalai model with RSI/momentum oscillators but I've never really looked at these and have no real idea how they work.
Attached File: SSRNid603481.pdf
Anything to add? 
my bank got pwnd 



schmitty


Total Posts: 57 
Joined: Jun 2006 


the + figure didn't have to be bigger than the  figure ...when it DOES oscillate (i.e. doesn't have runs) then rebalancing DOES make money
Let's take your +50/50 oscillation with a constant 25% proportional allocation. Start with a portfolio of $400. So your first bet is $100 (400 x 0.25). You win +50% = +$50 so now you have a portfolio of $450. Your new bet is $112.50 (450 x 0.25). You lose 50% = $56.25 so now you have a portfolio of $393.75, for a loss of $6.25 or 1.56% for the full period.
The end result is the same  a loss  for any constant proportion above zero. If the rest of the portfolio is in a riskfree asset, the Kelly optimal constant proportion is 0%. If you know that the stock will oscillate the optimal strategy is long on the upleg and short on the downleg. That would be 50% per period with no risk. Such returns do not exist in the real world.
Traders have been trying to fit cycles at least since JM Hurst published "The Profit Magic of Stock Transaction Timing." I don't know of anyone who has been successful at it.
I will read your links and get back to you on them.




NIP247


Total Posts: 544 
Joined: Feb 2005 


The price of one stock remains constant and the price of the other stock alternately halves and doubles
It might just be a misunderstanding from my part but the quote above does not imply 50/50, but rather +100%/50% or { 2x , 1/2x }

On your straddle, done on the puts, working the calls... 



schmitty


Total Posts: 57 
Joined: Jun 2006 


It might just be a misunderstanding from my part but the quote above does not imply 50/50, but rather +100%/50% or { 2x , 1/2x }
You're right, I missed that. Thanks. I should have read KR's reply to me more closely.
I was relying on KR's original definition in the post I originally preplied to:
That 'volatility pump'... i.e. the stock that oscillates between +50% and 50%
He also wrote in his reply to me:
the plus number does not have to larger than the minus number....[constant proportion] still makes money...
So you can see how I might have overlooked the correct numbers earlier in the post. The +100/50 numbers are the original Shannon's Demon numbers, I don't know where the +50/50 numbers came from.
I also stand by my comments about easily predicted/observed (and traded!) cycles not occurring in any realworld stock market. 



IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Hi schmitty,
Are you trying to tell us that it is not easy to make free money?!?!
By the way, welcome to NP! 




kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


OK, I stand corrected on the 50/50. The numbers there are simple so that it's easy to figure it out in your head. Anyhow, the point is that the risky stock doesn't need to go anywhere in order for the constantrebal portfolio to make money. My firstorder calcs show if the 'volatile' stock goes 1, 1x, 1, 1x, etc. then constantrebal annual return is better than x^2/8. This is estimate is not so good for bigger x  gives 3% when the true figure is more like 6%. But it IS greater than zero, even though the return on the individual stocks is zero.
Re: cycles, my idea traces to the simple model from Campbell/Lo/Mackinlay eqn 2.2.9. Without regard for the size of moves, you have a Markov model for market direction:
p(up / last move was up) = alpha p(down / last move was down) = beta
Obviously pure random walk has alpha=beta=1/2. The idea in CLK is to test whether markets trend or not  which is an obvious question since trend following is the first thing people talk about in technical trading. But if this model is not trending, and it's not randomwalking, then it is oscillating. Depending on alpha/beta it's easy to determine the optimal strategy  call that today's homework. In either case though, I don't really see why one would prefer the UP strategy, even with alphas and betas drifting around. 
my bank got pwnd 


toppy


Total Posts: 1 
Joined: May 2007 


kr, thanks for both articles. really appreciate this 




 

sammus


Total Posts: 148 
Joined: Jun 2004 

 

NIP247


Total Posts: 544 
Joined: Feb 2005 


Sammus, thanks a lot!

On your straddle, done on the puts, working the calls... 


Nonius

Founding Member Nonius Unbound

Total Posts: 12748 
Joined: Mar 2004 


Question on info theory. Suppose I have a finite alphabet and a noisy channel spitting out signals. So I have a string of letters a1....an and I want to compute the mutual information of that string with an+1. Now, suppose I abelianize the string, that is, I just write it as one letter to one power times another letter to another power etc. I then compute the mutual info of that modified language in which all letters commute. Is it true that the mutual information at least as good as the original case and possibly better, and better only if there IS non zero mutual info? 
Chiral is Tyler Durden 




I believe the reality is somewhere in between UP and constant rebalanced portfolios, is somebody working in a function which takes into account the trend and mean reversiĆ³n properties to swich from one algo to the other?
besides what do you think of ANTICOR & OLMAR ?? 



goldorak


Total Posts: 1050 
Joined: Nov 2004 


Both pieces of overfitted shit.

If you are not living on the edge you are taking up too much space. 




The closest I can think of is a short ATM put spread on SPY/SPX, where you collect 50% of the width of the strikes, and you bet 1 Kelly in it, keeping the rest in cash. The edge? The "drift" part of the "random walk with a drift" description of the equity indices in the US. Feel free to tear this idea apart, or use it to become the new billionaire philanthropist. Regards, TenDoublings 
Can I double my capital ten times in one year? 



Any thoughts on this? Or should I pick up my diapers and go home? 
Can I double my capital ten times in one year? 







