Page 1 of 2

Portfolio heat

Posted: Tue Dec 16, 2003 3:46 pm
by Chris67
Dear c.f. .. Thanks 4 an excellent forum.
I have 2 questions that have bugged me for years .. this is a journey I know ..
If the original turtles risked 1 % for n then by the time you are fully loaded in one product you have 5 n of risk on board = 5 %
That is effectively 5 % risk on one view ...
Say you had 3 fully loaded positions long and 3 fully loaded shorts .. thats 30 % of open portfolio heat .. especially if you are still dependant on 2n breaks and the time stop ( 10/20 days) hasnt come into play yet..
In my experience of raising money for a hedge fund most customers wont even tolerate 5 % of open heat and to me 30 % of open heat seems huge. Also 5 % risk per view seems large .. In november i had 20 back to back losing trades .. If i had used 1 n = 1 % i would be out of business .. this has happened many times in the last few years .. i find that even by using 1n = 0.2 % of capital p/l swings are large ??
Thanks in advance for your thoughts :D

Re: Portfolio heat

Posted: Tue Dec 16, 2003 4:10 pm
by TK
Chris67 wrote:to me 30 % of open heat seems huge.
One interesting opinion on portfolio heat appeared some time ago on Tharp's forum.
A guy who chose to be known as bojangles wrote:Another person, another opinion:

I trade a trendfollowing system similar (but not identical) to the one you're talking about. I trade about 30 commodities and limit my total portfolio heat to 52.5% of total account equity. Yes that is correct, fifty two point five percent. That means I trade in a way that Van Tharp and Ed Seykota strongly recommend against. But they're not me and I'm not them.

In backtesting this particular combination of System + Betsizing, the worst drawdown even seen at any moment between 1979 and today, was minus 40%. OF COURSE the future is not guaranteed to perfectly mimic the past, etc etc, blah blah.

My system and I were long all the energies and all the bonds starting in Dec/Jan. Made an enormous profit and gave it all back; presently with my HUGE GIGANTIC UNRECOMMENDED betting method, I'm in a 30% drawdown. No biggy.

You may feel differently.

BJ

Posted: Tue Dec 16, 2003 5:28 pm
by Forum Mgmnt
If the original turtles risked 1 % for n then by the time you are fully loaded in one product you have 5 n of risk on board = 5 % That is effectively 5 % risk on one view ...
Well, first I'd have to say that I made 110% for four years using that level of "heat", and I don't think many investors are expecting those type of returns or are prepared to take the level of risk we did. We didn't care about total equity drawdowns at all, only closed equity drawdowns.

If I were trading the turtle system by itself for a fund, I wouldn't trade more than 20% of the level we used to trade. That also happens to be the level you are trading.

I don't remember ever having more than 4 to 6 units in one direction when I was loaded in the other direction, so as a practical matter with a bet size 20% of the Turtle level, you'd go over 4% rarely and 5% almost never.

I personally don't think that adding positions in uncorrelated markets increases your risk. If the markets are truly uncorrelated, then adding new positions decreases your risk.

Posted: Wed Dec 17, 2003 3:20 am
by damian
To nobody and everybody:

I am not at all answering the question of this thread, but the below is related to the topic of portfolio heat from an open equity perspective in particular.

Open risk to paper PnL becomes an issue after fast moves in the direction of the trade. Steady trends (perhaps recent example is BO) do generate large open risk, but not absurd. Your broker may point out to you that your stop loss is more than a few limit moves away from the current market. He may think this is huge open risk but for a typical LTTF system that uses the lowest low of the last 50 days or an 80 day SMA for a stop, it is normal open risk. And generally everything turns out ok, except when there is a fast reversal against the position. This happened in March this year. I am yet to devise a good edge in dealing with these matters.

BUT, I have devised a potentially great edge in situations where the market moves very quickly in the direction of your trade. "Very quickly" needs to be defines. I have my definition and it all has to do with unit ATR moves compared to momentum. Do not worry too much about how I do it as there is more than one way to define "fast". If you are in a trade and your "fast" indicator signals then something must be done. I tighten my stop by a massive amount, I literally hug the market and often get out right in the middle of the blow off top. Great, what now? I sand paper my finger tip and get ready to break into the safe again: I get ready to re-enter on my re-entry rules. Lets say after 5 days I am signaled to re-enter and I do so. My new trailing stop loss is already tighter than it would be if I were still in the original trade, but not as tight as it was after the "fast" light came on.

Here is an example, but not what I do:

1. buy on a normal trend following entry - my trailing stop is 73SMA
2. 3 weeks into the trade the "fast" light flicks on.
3. I tighten my trailing stop to today’s low.
4. I am stopped out
5. I re-enter (a few ways this can be done, lets say at the high of the bar you were stopped out on) ** KEY QUESTION = what is my re-entry position size based upon???**
6. my trailing stop is now 20XMA
7. the trend continues
8. I am stopped out at 20XMA (or perhaps "fast" comes back on)
9a. I re-enter... this cycle could happen by a number of variations of sequence for quite some time.
9b. I never re-enter, but I am out of the trade and the market is well above 73SMA (my original trailing stop).

73SMA would have remained my original trailing stop had "fast" never come on. Once "fast" comes on, the 'trade management' that unfolds from then on becomes quite unknown. Normal, trade management (which determines open risk) is very certain: hold the trade and get out at 72SMA regardless of what happens. As I mentioned in 9a above - any number of exits an re-entries could take place, depending on what the market does and if "fast" shows up at the party again. I am trading BO in the same way that the turtle system would trade it. But I certainly did not trade the JY long in the same manner, nor the recent NG reversal. These trades are still live, but I do not have a position in either.... but I do have re-ntry stop orders in the market for both. Likewise Corn - just re-entered, but yet to hit my CT stop re-entry. The standard (and quite acceptable) trend following system would have maintained a position in these markets from day 1 till today. As a results open risk would have been an issue and so to would have volatility of returns.

Ok, so where is the meat:

- in backtesting, this concept is superb.
- in paper trading it has reduced volatility of returns by a great degree (relative to my LIVE trading).
- in real trading over the last month or so this concept has helped out quite a bit.... I look forward to the next year......

BUT: the meat depends on the answer to the ** KEY QUESTION **

As I wrote all of this I realised that I was leaving so many open questions and confusing statement. Happy to answer questions later - it is easier than anticipating all questions as I write.

Oh and also – one can apply this concept to any underlying LLTF system.

damian

Risk

Posted: Wed Dec 17, 2003 8:11 am
by kianti
There a different definitions for risk, I consider risk simply: entryprice - (stoploss price + slippage); I don't regard open position profits as risk and as a consequence I refer to the Closed Equity DD instead of the Open Equity DD.

Volatility of returns is not necesseraly a bad thing. According to Graham Capital Management:
[quote]
A trend follower achieves positive returns by correctly targeting market direction and minimizing the cost of this
portfolio. Thus, while trend following is sometimes referred to as being “long volatilityâ€

Posted: Wed Dec 17, 2003 8:54 am
by Sebastian
Damian, my response to the "key question" of what position size when you re-enter: Treat the trade as a "fresh" one (ignoring the fact that you got out of the last one on a dramatically tighter stop) and size accordingly. Here's the reason I say this.

I was running some tests on my ersatz Turtle system in Excel and wanted to study the losses to see if there was some way I could reduce them. What I EXPECTED to see was that the largest losses would come from situations where there were full-boat positions and the market sharply reversed trend.

What I found instead was a real eye-opener. The largest losses came from relatively SMALL positions where the market suddenly reversed trend. In thinking about it, this seems sensible to me when you consider the nature of trends. They rarely make spike tops/bottoms and totally reverse direction without any warning. Mostly when they get ready to turn they exhibit a sharp increase in volatility in the opposite direction.

If you've built up a large position during a long trend move, you'd get stopped-out of it on a sharp move in the opposite direction that could be the pre-cursor of a total reversal of trend. Re-building the position anew should result in a smaller position than you had before since the trend is almost over and won't run far enough to trigger all the entries, a good thing.:)

On the other hand, if your big position got stopped-out on a transitory "exogenous event" of some kind that doesn't change the long-term trend, when you re-enter it will move persistently enough to trigger all your entries so that you can go full-boat again.

FWIW.


Luck,

Sebastian

Posted: Wed Dec 17, 2003 9:17 am
by -w.
damian wrote:** KEY QUESTION = what is my re-entry position size based upon???**
It really is the key question, isn't it? While I find the algorithm you're using interesting, what it basically boils down to is it's second-guessing the trend, IMHO. You're describing the typical trendfollower's dilemma: If you re-enter the position at any size smaller than the closed-out size, you have effectively limited your profit potential, a No-No as I understand it. If you re-enter the position at the full size you wouldn't have needed to close out in the first place.
kianti wrote:I don't regard open position profits as risk and as a consequence I refer to the Closed Equity DD instead of the Open Equity DD
I totally second that. c.f.' view to "simply" reduce size to keep Open Equity DD in check seems to be the most sensible solution when dealing with OPM. It also has a built-in additional advantage: You can trade the same system as your clients without having to give up performance if your personal risk tolerance is higher. Greatly helps confidence, I think.

-wojo

Posted: Fri Feb 13, 2004 9:53 am
by King_Tiger
I'm new to website and have a question re total risk exposure.

I understand that each bet should not be risking more than 1 or 2% of your equity.

My questions are:-
1) Each risking 2% and if you have 10 positions, that's like 20% risk exposure to your equity. So how to manage total risk when you have several bets/positions in your portfolio?

2) Several months ago, I had the unfortunate experience of seeing one of my stocks fell by 98% at the open. I haven't sold this one out because 1) it's not worth anything now and 2) I want to remind myself that there is always a risk that any one of my stocks can drop to zero tomorrow morning.

But if it's not a gap down of 98%, just 10-30%, do you wait until the price comes back up a bit before selling or you just sell it straight away? From c.f.' rules in www.originalturtles.org, it seems the answer is to wait a bit. But you never know if the price will come back up, right? What if it continues to drop?

What can we do this situation?


3) I have a portfolio of stocks worth $240K supported by an equity of $200K. And I want to trade FX (lots size can be 20K+). Can I use the same $200K to support my FX trading? That is, 2% of 200K per FX bet, etc.

Any comments are welcome and thanks!

Posted: Fri Feb 13, 2004 10:57 am
by shakyamuni
Forum Mgmnt wrote: I personally don't think that adding positions in uncorrelated markets increases your risk. If the markets are truly uncorrelated, then adding new positions decreases your risk.

I agree.

Posted: Fri Feb 13, 2004 11:14 am
by gms
King_Tiger wrote:1) Each risking 2% and if you have 10 positions, that's like 20% risk exposure to your equity. So how to manage total risk when you have several bets/positions in your portfolio?
I'm don't trade according to "turtle rules", so I can't give you the official c.f. the Tortoise answer :lol: , but a way to manage risk is to invest/trade in uncorrelated markets, and also to think in terms of allocations of your portfolio to cash and index funds. In your stocks, stops manage risk, diversification manages risk to some degree. There is diversification in stocks if you allocate to different sectors, and in different stocks. But for stocks, the diversification ends where the general markets' influences begin.
King_Tiger wrote:2) Several months ago, I had the unfortunate experience of seeing one of my stocks fell by 98% at the open. I haven't sold this one out because 1) it's not worth anything now and 2) I want to remind myself that there is always a risk that any one of my stocks can drop to zero tomorrow morning. But if it's not a gap down of 98%, just 10-30%, do you wait until the price comes back up a bit before selling or you just sell it straight away? From c.f.' rules in www.originalturtles.org, it seems the answer is to wait a bit. But you never know if the price will come back up, right? What if it continues to drop? What can we do this situation?
Are you trading penny stocks? That 98% plummet would make me wonder if my portfolio selections are off. As far as whether waiting for a bit of recovery or not, this can be answered in 2 steps. First, if the stock is of questionable value and there is therefore good reason for the decline, then either you should be short instead of long or you should change your portfolio criteria. If it's a strong stock, but fast spiking down on huge volume because of some recent sentiment against it that isn't of the nature to devastate a company, you may look at that activity as fear gripping the investors. Odds are that when the smoke clears and the stock finds support, it will be bidded up. Perhaps not as fast as it went down, but certainly a bounce up after a couple of days will get you out in a better position than if you atempt to get out at the lows of the spike. What if it continues to drop? That's the chance you take. There's one type of decline that may recover, there's another that may not. To be able to better assess the situation, you have to look at volume levels, time, previous S&R and the price, as I've suggested above.

Now, this is all discretionary trading. If you have a system that you need to follow rules of, then discretionary trading may skew your system.

Correlation

Posted: Sat Mar 20, 2004 7:01 pm
by pruggera
Forum Mgmnt wrote: I personally don't think that adding positions in uncorrelated markets increases your risk. If the markets are truly uncorrelated, then adding new positions decreases your risk.
I notice that when correlation is mentioned in postings to this forum it usually is about the correlation between markets. This may be easier to obtain, but don't we really want to know the correlation that occurs in markets only when they are both in a trade? We are only impacted by correlation DURING the trade. What does it matter what a market is doing while it has not been selected by the trading system. In fact, does including the time when both markets are not in a trade reduce the validity of the resulting correlation study?

Phil

Turtle Correlation

Posted: Sat Mar 20, 2004 8:07 pm
by ksberg
Hey Phil, good to see you here! :D
This may be easier to obtain, but don't we really want to know the correlation that occurs in markets only when they are both in a trade? We are only impacted by correlation DURING the trade. What does it matter what a market is doing while it has not been selected by the trading system. In fact, does including the time when both markets are not in a trade reduce the validity of the resulting correlation study?
There are two distinct aspects going on here: (a) correlation between two time series, and (b) how we apply correlation information to trading. In the first instance we can talk about what we're correlating (markets, market segments, total equity, risk, etc). But the second aspect is at least as important as the first, and it starts to address your question.

For instance, if we were applying Modern Portfolio Theory (MPT) we might use correlation to select and balance markets as a part of our portfolio such that the value of correlation pairs are minimized. If we had already selected a basket of markets, we might use correlation as an input for limiting portfolio heat, such that we block taking new positions that are highly correlated. Or, we could use correlation to coordinate how stops are managed across related markets.

Turtle trading uses correlation to limit portfolio heat, blocking trades after a maximum unit threshold is reached. When you think about it, Turtle rules only consider correlation when in a trade, just like your comment.

Cheers,

Kevin

Re: Turtle Correlation

Posted: Sat Mar 20, 2004 8:37 pm
by pruggera
ksberg wrote:Turtle trading uses correlation to limit portfolio heat, blocking trades after a maximum unit threshold is reached. When you think about it, Turtle rules only consider correlation when in a trade, just like your comment.
Yes, Turtle rules wisely use correlation information only when in a trade. My concern, however, centers on how that data is developed. The approach taken to develop the correlation data can have a large impact on the actions taking by the trading system. If non-trade time were removed from the correlation calculation what is considered correlated and not correlated could be different. Or it could make no material difference.

Phil

How data is correlated

Posted: Sat Mar 20, 2004 9:44 pm
by ksberg
Phil, I guess my response is ... well, ok, what do the results say?

Puzzling Correlation

Posted: Mon Mar 22, 2004 10:31 am
by lperepol
I have been kicking the can around about correlation. My concern is that correlation is backward looking and thus is use for trading in the “now “is limited. Any type of diversification is a good thing.

Say we have two time series A and B.

Then we construct a table

<table ALIGN=â€

Posted: Mon Mar 22, 2004 1:32 pm
by Hiramhon
It's not so terribly complicated; each instrument can be in one of exactly three states (long, short, out), so for a pair of instruments there are only nine cases to analyze (3 states for A, times 3 states for B, equals nine cases). Your table lists eight of the nine cases.

When you write computer software to perform these analyses, a mere factor of nine additional CPU-time will be negligible.

However, I think stancramer's partitioning method is perhaps not so fruitful. I suspect it would be far more beneficial to the trader, to partition the equity curves into (a) runups; (b) drawdowns; and (c) choppytimes. I don't know about you, but I am REALLY interested in the correlation of B's equity curve to A's equity curve, during those periods when A's equity curve is in a drawdown. These are the times I'm really hoping for some non-correlation :wink: . But of course you are free to write your software however you wish, heeding your own counsel, and to gleefully disregard my observations if you want.

Posted: Mon Mar 22, 2004 10:04 pm
by lperepol
I am not sure if it would be that simple if you look at a portfolio. I am thinking in terms of combinations of n distinct instruments of one unit.

C(n,r) = n!/(r!(n-r)!

For arbitrary r the combinations increase and increase more if one allows pyramiding of instrument units. In addition, the resultant combinations would have to be multiplied by two, since each instrument can transition to a long or short state.

And now use correlations to pick the best of the lot and see if a correlation does indeed make a significant difference. For example compare combinations chosen with rules of correlation with combinations chosen at random. In other words compare pure diversification with optimized diversification. Pure diversification works well for me and it is simple. Correlation raises a multitude of hypothesizes that can be tested.

I am just thinking out loud. However, I do think I need to think some more.

Simplicity?

Posted: Mon Mar 22, 2004 11:47 pm
by ksberg
Since Phil broached the subject I've given this some thought, and even managed to implement a few accomodating extensions. So I have to say the exercise has been worthwhile so far.

However, I'm really beginning to doubt the premise of "only on" correlation. Why? Consider the type of trade when one market/system is winning vs. another is losing. At least in my systems, losers are not held very long, while winners are just the opposite. That means that the "ideal" low correlation will involve lots of small losing segments mapped onto a winning segment. Now, instead of correlating two markets we're correlating thin pieces of markets stitched together via a two input functions.

Instead of piece-wise segments, probably the most straight forward thing to do is correlate equity streams between market/systems. This automatically accounts for when we're in and out of positions and we don't have to do segment gymnastics.

I didn't mention it before, but what I've found is that correlation values appear more consistent on higher time frames (weekly, monthly) than daily. That is, I can count on weekly correlation readings but daily more often moves from being correlated to un-correlated. It's unproven, but doing short piece-wise segements is likely to vary even more than dailies.

Thoughts?

Kevin

Omit this is message , it is erroeneous beyond repair!!!

Posted: Tue Mar 23, 2004 8:14 am
by lperepol
Omit this is message , it is erroeneous beyond repair!!!


I looking for comments on alternative objective functions of correlation and diversification.

Here is some of my reasoning. (Omit this is faliciuos)
Take two instruments A and B.
such that correlation(A, B) = 0
We now have zero risk and a zero utility growth rate.

The conventional objective function of diversification is to reduce risk. However, I believe that a side effect of this type of an objective function also reduces a utility functions growth rate – no profit.

I think a better objective function would reduce risk and increase utility growth rate.

In order to increase utility growth rate the long instruments in a portfolio would have to be highly correlated and the short positions would have to be highly correlated in a negative sense with longs on average (moving up fast) and shorts on average moving down fast.

Logically, if one sums correlations of pairs the instruments and arrives at a zero or near zero correlated portfolio then the portfolio will have poor performance and low risk. The value of such a portfolio would have questionable value – except for that fact that one is participating in trading and investing just for the sake of belonging to a financial community.

Comments?

Posted: Tue Mar 23, 2004 11:37 am
by Hiramhon
Take a long position on A and a short position on A.
Now we have a zero correlation:
correlation(long(A), short(A)) =0
That isn't true. The correlation of long(A) with short(A) is -1.00000. Try it in Microsoft Excel, the function name is CORREL. You'll get negative one. Honest. Its online help is cut-and-pasted below.

A correlation of +1.0000 means whenever X goes up, Y also goes up. A correlation of -1.0000 means whenever X goes up, Y goes down.

A correlation of zero means Y's behavior is completely independent of X. Sometimes when X goes up, Y goes up. Other times when X goes up, Y goes down. And other times when X goes up, Y does nothing.

CORREL:
Returns the correlation coefficient of the array1 and array2 cell ranges. Use the correlation coefficient to determine the relationship between two properties. For example, you can examine the relationship between a location's average temperature and the use of air conditioners.

Syntax:
CORREL(array1,array2)
Array1 is a cell range of values.
Array2 is a second cell range of values.