|
Diamondjag
Posts: 123
Joined: 10/11/2012
Location: Brighton, Co.
 User Profile |
A comment by Steve Mayo on yesterdays OmniVest round table made a big impression on me. In Ed's notes today, he interpreted Steve's comments as, "Some of the input was especially powerful. Steve Mayo talked about his experiments, turning Portfolios ON and OFF based on simple examinations of the equity curves. I believe he said he doubled the results in his simulation by doing this."
What I heard Steve say on the call was that his dramatic OV improvement came from switching to the best performing portfolio at the first of each month based on a moving average (I assume of the portfolios equity curve).
In another post Steve said, "But, perhaps like classic technical analysis, it seems to me to be far more important to have automation that could switch to any higher-performing portfolio at the optimum time, rather than focusing effort on building the optimum portfolio to begin with. IMHO, simply (?) adding basic trendline or momentum indicators on the equity curve and supporting the ability to do automated port switching looks to be a far better near-term goal."
I have an interesting example of portfolio switching. I use a program called "Dynamic Investor Pro" (the only other product I use other than Nirvana products). It is basically a relative strength program, keeping you in what is the strongest stock/ETF/or Mutual fund at any one time. You have your choice of different relative strength formulas. What I have found is even using the same group of symbols, different relative strength formulas give you vastly different returns in different years, regardless of the overall market return.
Example:
Using a relative strength formula called Price Oscillator with a mix of five non-correlated ETF's, the portfolio has already returned 18% this year, with the market down about -.4% (prior to today). Last year those same ETF's and same formula returned a -34% with the market up +29%. In 2010 they returned +36% with the market up just 13%. This is just one of many, many examples. I am now switching formulas in this program as returns change. You clearly can't depend on one relative strength formula holding up in all markets. I look at results weekly but may only switch once is several months or less depending on how things are going. Some portfolios remain the best all year long (only to fall apart the next year). And, you may have to totally stop trading and move to a different mix if none of the relative strength formulas are working with that certain ETF (or stock) mixture.
The point....it is possible that we'll see the same with OmniVest, as Steve seems to be finding. For optimal returns, you may have to keep moving....staying in the Portfolios that are working at any one time (and this is not necessarily long to short portfolio switching). Switching can clearly get carried away by changing too often, but some portfolio switching could dramatically increase returns and...wouldn't it be wonderful if OmniVest could monitor the portfolios and do the switching for you.
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
Good thread topic. I've got some additional thots.
1. Steve's mechanism forced switching by closing all trades before switch. With OVest and TProc running the show, I believe that active trades in closing portfolio could/should be allowed to run their course controlled by their TPlans. Any new trades would come from the newly selected portfolio.
2. The gotcha of filtering/conditions of virtually any sort is that the filter is too sensitive and it whipsaws, or it's stable but lagging. Finding a balance is tricky. As I've presented elsewhere at length, I think that at least six and more likely nine market states are useful - bull flat bear trends, each with wild normal or quiet volatility. Hopefully the mechanism for condition switching can somehow work with this multistate blend. Maybe "blend" is an important concept to work with. Someone (Ed,maybe?) suggested using the OScript formula to determine a balance or weighting of portfolios, rather than a square wave shift between them.
3. So, here's an idea: allow an OScript formula to be assigned to each of several portfolios, to act as a weighting factor for the allocations in that portfolio. Example: three portfolios, one designed for clear bull trends, another for clear bear trends, and a third for transitional/flat/consolidation periods - each with its own OScript formula that creates a numeric weighting factor based on some (presumably) simple market-metric, geared to recognize conditions that "fit" it's related portfolio best. Any of the three formulae could create weightings from 0-100%. That weighting would be applied to the allocation percentages in the various strats in each portfolio, normalizing the results to the desired pct invested for that Account. OVest already does just that with strats combined into a portfolio.
The result would be an Account that intelligently "glides" through trend changes, in a manner that permits simultaneous mixing of portfolios in situations where the transition is not cut and dried. In my belief, that is the norm and not the exception.
One could take this further, with three additional portfolios for wild normal calm, and OScript weights for them.
Or, some entirely different track - nine portfolios for nine Sectors with OScript weightings gliding the mix through group rotations.
4. How to do this in the interface to make it simple to understand without adding a gazillion more layers: I believe the natural solution is to allow the user to create a portfolio, then STORE it AS a strategy, so that it as a whole can be studied and manipulated using the same very cool SLab tools that now work with strategies. And since strategies already have allocation percentages and conditions and metrics and whatnot, this might allow the portfolio-blending mechanism to be implemented with relatively small impacts on the GUI.
Maybe I should have said "system" not "strategy" - but as I understand it, in OVest, "System" = what OT calls a strategy, and OVest "Strategy" = OVest System plus List plus Direction plus Conditions - if that's true then Nirvana please write it down prominently somewhere. I hope I got it right ;-)
|
|
smartprofit
Posts: 14
Joined: 3/2/2014
User Profile |
Interesting post! Was that round table recorded? I had to work that day.
|
|
Diamondjag
Posts: 123
Joined: 10/11/2012
Location: Brighton, Co.
User Profile |
I don't believe so. As I remember, Ed said he didn't think about recording it until it was to late to set up the recording.
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Good thread… portfolio switching works.
Steve Mayo & I have been working intensively on ways to improve the already good results with OV over the past year – well north of 1,000 hours devoted to development and testing of different theories. We’ve learned a lot - about OV and from each other - and found the best results in the area of weighting some “proprietary” parameters and then using those weights to choose which portfolio to use for the following month (switching). The bad news is that my archive of Excel spreadsheets that didn’t work contains over 6,000 “development” (i.e. didn’t work) spreadsheets (70GB), and my current portfolio list in OV has more than 300 portfolios (250 already deleted). In addition, the macros in the spreadsheets that are currently producing results can take many hours to run.
I guess what I’m saying is that this level of testing isn’t for the faint of heart, but I FULLY agree with Ed’s plan to have the Professional version of OV for those that enjoy the development aspect. I’m convinced the returns are worth the effort, and hope that Elite trader will provide a way for many of us in this forum to share (and profit from) each others work.
I’ll post a chart below to support Ed’s commitment to providing more tools – OV has capabilities we’re only beginning to appreciate, and the new tools coming can only improve the results.
These are the results of running weighting algorithms at the end of each month to select a portfolio to trade the next month from 1/1/2007 to 1/1/2014. Ed’s excellent ARM4 Margin portfolio has been our “benchmark to beat”, so that’s included along with the QQQ @ 2X margin.
In this example, our algorithm switched between two optimized portfolios on 36 of the 85 months (42% of the time), simply “carrying over” the preceding month’s portfolio for 49 of the months (58% of the time, no portfolio switch). In this example, it was accurate 65% of the time in picking the best portfolio to trade for the forthcoming month, improving CAGR by 195% and the Return:Risk Ratio (7yr CAGR/7-yr monthly SD, sort of a modified Sharpe Ratio) by 26% from 7.0 to 8.8.
Steve & I have debated about posting results like this without wanting to divulge the specifics, but we feel it’s appropriate now as a small indication of what’s currently possible - and to support Ed and the staff in their complex and difficult move to another, even better level with the new tools coming on line.
BTW – Jim’s post above has pretty much touched on all the correct aspects we’ve found to be useful (right once again, Jim).
Mark
 Attached file : Port_Switching.jpg (213KB - 1407 downloads)
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Nice thread!
My observation and suggestion, now that we know portfolio switching can greatly improve returns, is to bake that knowledge into OV so all users get the benefit. Simple, on/off switch. Maybe a drop down to select between 3 - 5 proven algorithms. Nothing more.
The Pro version sounds awesome. I will probably sign up, even if I never use it, just to support Ed's team. But I strongly believe it's in everyone's interest to bake proven concepts into the base product. AND EXCLUDE unproven stuff that adds to the probability of user failure (as well as complexity) for non-Pro users.
I feel the interative strategy selection (ISS) method of portfolio creation falls into this camp. Many hours of manual testing have proven the concept (IMO). The base user could get a simple interface to turn the feature on/off, specify the depth (number of strategies), and the metric to solve for (like max return vs min risk). Very simple. Pro user could code up more complex metrics and strategy selection criteria.
Both portfolio switching and ISS could be accused of "curve fitting". I would argue all technical analysis is curve fitting. But using broad concepts that "make sense" and have held up to back and forward testing with "out of sample" data offer far more promise than throwing darts or randomly tweaking a large multitude of parameters which aren't really understood by the typical user.
How about using ARM to decide which indicators or fundamentals best determine portfolio selection? How about coding up the process that Steve and Mark have been laboring over such that the computer works hard and we all reap the rewards?
Cheers
Keith
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
BTW - which portfolios are you switching between? How were they created? Would better portfolios enhance the returns generated by portfolio switching?
Keith
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
Just like individual stocks and mutual funds, strats and ports alternate between trends and trading ranges, and (hopefully) they do so at different times. Rather than trying to build the "optimum" portfolio for all market conditions, this approach instead tries to identify a better performing portfoliio and switch to it for the forthcoming month. Rather than starting with strategies, it simply starts with portfolios, which already have much better reward:risk metrics compared to individual strats.
Mark and I should point out that the example above uses a more sophisticated ranking process than what I had suggested to Ed in the survey, that being the ability to rank and switch between portfolios (and ideally strategies too) based on a simple indicator measured on their equity curves.
Therefore, here's a second, less sophisticated, example along those lines. It switches 7 times between a pair of portfolios to use the portfolio for the forthcoming month that had the higher monthly return in the preceding month. Just this relatively easy approach increases the modified-Sharpe to 8.5, about 20% better than ARM4 Margin over the same period.
This simple technique could theoretically be done today by any of us simply by running OV simulations each month to rank the monthly return. But, be aware, as Jim points out, that this Excel simulation does assume completely closing all open trades in the first portfolio and buying all open trades in the second portfolio on those 7 month-ends where a switch occurs (and it doesn't include the added commission costs for doing so).
Unless you were to close all positions manually, OV would normally stay in the open trades from the prior portfolio until each position reached its sell point. Without Nirvana adding functionality to OV, we can't know what impact those carry-over trades might have. Given the short duration of the RTM trades, it would logically be better to NOT enter those positions mid-trade just because the calendar cliked off a new month. I THINK the results would be better than this simulation, but currently there is no way to know for sure.
Also, our simulation is based on the mark-to-market data from the OV graphs, which is the equity sum of the positions at the end of the day; conversely OV makes its trades at the open. Our simulation assumes you could sell-out/buy-back-in at the market closing price at month-end. Given OV doesn't tell you what it intends to buy tomorrow until well after market close, you obviously cannot actually do that. The simulation assume you would buy the next month's positions as of market close, even if they were going to be closed out in the morning. In reality, you would close out at EOD and then buy back in the next morning at MOO but that's an overnight slippage that is then also not included in this simulation.
Bottom line, it certainly looks like there is good potential here to improve upon the OV returns but unless/until Nirvana can add the ability to rank-order ports (and hopefully strats too) based on an indicator, and then to trade in/out and between those ports/strats based on that ranking, we won't be able to properly validate that potential.
Attached file : PastedGraphic-26.png (47KB - 1369 downloads)
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
Portfolio optimization is a much more difficult objective. Determing the best set of 5 out of 100 choices takes 75 million permutations. Crank that up to the best-10 portfolio and you jump to 17 trillion permutations! And now with v2, we have a lot more than 100 strats to choose from.
Iteratively adding strategies to a mix (ISS) sounds like a promsing way to reduce the number of permutations, but you have to understand how OV works to see why it may not be. When you add the next strategy to the current mix, OV is now presented with a larger set of candidate trades to take each day. OV now might take a trade from this newly-added strategy and then not have the equity available to take the trade it took from the original set of strats. It's not additive/subtractive, it's different. (Statistically speaking, there is a high degree of nonlinearity.) Which means it is essentially unpredictable and we are back to doing all those permutations again.
Conversely, portfolio switching requires only adding the ability to calculate a single formula that ranks a few candidate portfolios and a trading rule that switches between (or out of) them based on that ranking. If you start off with some fairly good portfolios (and Nirvana is promising us lots of them!) then your downside is limited to just the worst of the best. When switching, say, between 2 good portfolios, as long as your algorithm accurately switches to the better portfolio for the coming month more than 50% of the time, you should do better than trading one of those ports individually (yes, I know ordering of wins/loses matters but that's a different post). Mark and I have now shown two examples that are accurate more than 50% of the time, and it didn't take 17 trillion calculations.
Plus, portfolio switching is an easy concept for novice traders to understand and could be applied to switching between Nirvana-created portfolios. The user would not have to know how to write a formula -- Nirvana could do that and then simply let users trade that "system of portfolios" as an alternative to picking just one.
Rather than devolve into an ISS debate, let's move that to another message string and keep this one focused on portfolio switching. Thanks!
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Ed was kind enough to take time out of his busy schedule today to respond to me and asked me to add this to the discussion...
I quote Ed:
"We are implementing the ability to turn Portfolios ON and OFF with simple OmniScript functions, and the Equity Curve will be available as a "symbol", just like any other, so people can use Moving Average of the Equity Curve to filter it."
More good news - this should create a lot of new possibilities.
Mark
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
yippee!
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Steve,
The current OV logic allows one to move between portfolios gracefully. The existing trades are managed to completion. New trades are generated from the newly selected portfolio(s). And it doesn't need to be a binary choice.
Adding the ability to adjust the Alloc % of various selected portfolios based on a metric would provide a lot of flexibility. If the metric was dynamically calculated and applied during the simulation period one would easily be able to quantify the benefit of portfolio switching. (Although perhaps not based on a monthly period.)
I think portfolio switching (actually rebalancing, which could include a complete switch between portfolios) is an excellent approach. It adapts to changing market conditions.
Assuming Nirvana comes up with a bunch of portfolios that perform good-to-great in every market as a starting point... I agree - one won't do worse than the worst portfolio, and probably will do better than the worst by portfolio switching.
Iterative Strategy Selection (hence ISS, not ITT) is a way to construct optimal portfolios given a specific set of account settings. ISS is not in contradiction to portfolio switching (which I'm getting tired of typing so I'll refer to as PPS for "programmatic portfolio switching"). PPS and ISS are highly complimentary. ISS gets one to portfolios that worked well during different markets. PSS allows for the automated (programmatic) switching (or rebalancing) between those portfolios.
Both allow software to perform analysis that would take a mortal many man hours to complete. Both are instances where the computer can do it better than a human. Both rely on hindsight. Both are better than throwing darts. (Much better IMO.)
I didn't mean to "devolve into an ITT debate". Sorry.
Cheers
Keith
|
|
Diamondjag
Posts: 123
Joined: 10/11/2012
Location: Brighton, Co.
User Profile |
What we're talking about here is essentially market timing. I know those are dirty words in most "investing" circles. I would just like to caution those thinking about timing a long/short strategy...going into one strategy when the market is going "up" and another when the market is going "down".
The first big question is when does the market turn? Then...how long is that turn going to last? We might be able to quantify the first (moving average cross etc.) but we will never know the second. More importantly, all "up" markets and "down" markets are not the same. Ed alluded to this in a post when he said," Reversion to Mean strategies (which is what all the "R" strategies are), work best in choppy markets that move up and down to provide reversion opportunities. In strong trends, Reversion Strategies will buy in downtrends and sell in an uptrend." Furthermore, a straight up market would work best with one strategy in OmniVest, a sawtooth up market (still an up market) would work best with another strategy, a swinging up market would work best with another strategy. So different strategies work best with different TYPES of up, down, and sideways markets.
This may not be a dramatic revelation to anyone but I would suggest, if you're not going to use a single Portfolio (I'm going to switch to the word Portrolio here since that's generally what we're trading in OmniVest) highly tuned over a long period of different market conditions, that you don't go for just an "up" Portfolio and a "down" Portfolio but keep a stable of good Portfolios, as this thread is suggesting, and move into them as they are working (defined as making money). Let the Portfolios themselves tell you which one to use at any one time.
|
|
BrianD
Posts: 302
Joined: 2/23/2013
Location: Grand Rapids, MI
User Profile |
Personal experience found switching between, or starting from, new portfolios in real time seemed it may take a few days for account to 'get up to speed', if you will. Most of my portfolios depend on high % vested to achieve targets, so one could expect.
My concern is a few days lag in month by month rollovers may make a noticeable difference in simulation performance metrics if not accounted for.
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
IMHO there is no benefit to sudden changeover. A blended shift makes more sense to me. The prior post is yet another reason.
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
I'm with Jim. I've seen many RTM trades go from loss to profit. Forcing a close of all trades locks in (realizes) losses that have a decent chance of turning into profits if managed to completion.
The devil is always in the details. Will OScript need to resolve to a binary result that enables/disables a portfolio? Will OScript be used to modify the Alloc % of a strategy? Will OScript for one portfolio's "condition" be able to reference the equity curve of another portfolio? Will one OScript be able to enable/disable arbitrary portfolios? And set allocation levels? Or will there be one OScript that runs every day that manages all portfolios?
What Steve has described (afaik) is -
if (end of month)
{
__find best portfolio return for past month
__enable best portfolio
__disable all other portfolios
}
This logic requires reference and control of all portfolios. I.E. a global scope, vs a private scope. (relative to a single portfolio).
It will be interesting to see what Nirvana engineers provide.
Cheers
Keith
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Also, moreover, what Steve is saying (afaik) is -
if end of month
{
__run each portfolio at 100% alloc, as the solely enabled portfolio, based on the account settings but with a date range that covers the previous month.
__find the portfolio with the best return over the past month
__enable the best portfolio
__disable all other portfolios
}
This seems like more than a OScript condition applied to each portfolio.
An alternative might be -
__run each portfolio with no regard to account settings, taking every trade as sized by the strategy and strategy Alloc %, with a date range that covers the previous month.
The latter might ameliorate some of the concerns Steve has voiced about Apple and Amazon hogging trades and skewing performance stats.
(Revisit having the engineers figure out what works best and provide an elegant solution...)
Cheers
Keith
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
Most of our collective knowledge comes from making one-off trade decisions. OV is making those too, but against the dynamic of a fluctuating available equity. That adds a new dimension that could change many "rules" we've all come to believe.
So, I also THINK that transitioning from one port to the next is probably best, but as a scientist I recognize that I really don't know until I can properly test it. Ed said he would be giving us the ability to export the "discarded" trades, so maybe at that point we can answer that question.
|
|
Mark S
Posts: 56
Joined: 10/11/2012
Location: Barrington, IL
User Profile |
I can't hold a candle to the level of analysis and back-testing that many of you discuss on these forums (thank you!), but here's a poor man's attempt to get at the benefits discussed in this thread, without having to wait for new code to be developed.
I built 4 portfolios, optimized for different market conditions, through heavy use of Steve and Ian's Portfolio Analysis tool (thank you!)
Bear
Bull RTM Strategies
Bull Trending Strategies (longer holding periods)
Both Bull and Bear (RTM Strategies that were common to Bear and Bull RTM above)
I then created an account that contains all 4 Portfolios (18 Strategies), and simply set the Allocation % to zero for the portfolio(s) that I think are not relevant for whatever reason. For example, I think we're due for a major correction, so I've set Bull Trending to 0%, and I've turned on the Bear portfolio.
The benefit (I think) is that any open positions will still be managed by the relevant strategy, but no new positions will be opened when a Portfolio is set to 0% allocation. I just started doing this last weekend, so have not yet turned off a Portfolio that has open positions. But I think this will work, and will facilitate easing in and out of strategies as needed or desired.
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
That sounds like an excellent idea Mark...
OV is supposed to handle the open orders (close per strategy) when you change portfolios, but if you're just "rotating" back & forth between a few portfolios, yours seems like an excellent way to do it. It might accommodate Jim's idea of "slowly"(?) rotating in and out of market states too if you found a good way to assign weights to them as the market changes...???
Thanks for the input,
Mark
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Earlier in this thread I posted a chart comparing the results of a switching algorithm Steve & I developed vs Ed’s excellent ARM4 portfolio.
It might be assumed we were switching between two “better” portfolios to achieve that performance relative to ARM4 – and that would overlook the inherent benefits of switching.
To visually demonstrate the power of switching, I’ve put together a series of charts using some standard account settings with ARM4 and another portfolio, both in OV and then using our switching algorithm.
Results in OV2;
Results of switching the same two portfolios;
I hope this helps everyone appreciate the considerable benefits of portfolio switching.
This was a long and tedious process the past year for Steve & I using Excel. We didn’t want to just “put this out there to brag” and then not give any details - but Ed has mentioned that this capability could be a part of the new upgrades coming to OV2. If so, it will definitely raise the potential of OV to new levels for everyone…
Mark
Attached file : Combo In OV2.jpg (540KB - 1061 downloads)
Attached file : Combo Switching.jpg (231KB - 1040 downloads)
|
|
Fred Gordon
Posts: 481
Joined: 10/11/2012
Location: Fayetteville, Ga
User Profile |
Mark, I thank you for your effort. your's is a very clear description of the benefit of an algorithm to facilitate portfolio switching. It would certainly raise "hands off" trading automation to a new level.
If Nirvana created such an algo and made it available to all, would'nt we tend to trade the same symbols at the same time?
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
Nirvana is not planning on creating the specific mechanism that Mark and Steve have worked so hard to develop. Rather, N is providing a means for Mark and Steve to implement the special rule that they have developed, on an automated basis, without having to check their Excel sheet every month.
OVest does now have and will grow further in its ability to help select the most effective combo of strategies and allocations to make up a good Portfolio (like M&S's Portfolio 2). M&S have automated that process, but automation like that will only be possible via the not-yet-clarified capabilities of the Pro level.
The most important thing for folks like M&S who have worked so hard to define good portfolios, is to be able to implement the switching on an automated basis. Assuming that the rule that you want to use can be coded into an OmniScript formula, the "input field" for that formula is what N is providing for Portfolio switching.
Of course, it's possible that N will eventually create an engine which does the kind of thing that M&S have been doing in Excel ... but afaik that has not been promised yet.
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
Additionally, in response to your post ...
Ed already has said they have plans for "intelligence" (ie some extra rules) to be added which will help prevent or ameliprate the effects of lots of people trading the same symbols at once. My guess is that as the OV clientele grows, there will be more and more work done on this issue.
Another note ... if M&S choose to make their custom setup available thru Elite Trader, then there could be a lot of people who trade it. But if they keep it to themselves, they won't likely have to worry as much about that.
|
|
Steve2
Posts: 750
Joined: 10/11/2012
Location: Annapolis, MD
User Profile |
M&S, nice work!
When each switch occurred, did you force-close open positions or allow them to be closed by the strategies in the "switched-from" portfolio? Are you able to tell from your data how much of the increased returns from switched portfolios vs combined portfolios were due to more profitable trades vs a higher overall TPM vs a higher average % invested?
Steve
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Excellent work and presentation, Mark and Steve! Thanks!
Personally, I don't mind having to check the performance of N portfolios at month end and then changing the portfolio I'm trading in my live account. Automating the switch would be nice, though...
What gets me is the work you had to go through to get your data. And because it is sooo painful to do such research it really thwarts further investigation, such as -
1) What works best? bi-weekly, monthly, quarterly, semi-annual switching?
2) Would a market criteria better determine switching vs. time constants?
3) Would rebalancing (vs binary switching) produce smoother equity curves?
4) How would solving for minimum time between new equity highs work out?
5) How about switching between portfolios specifically tuned for differing market conditions fair?
6) Does switching hold up with any set of portfolios, or are these findings one-offs?
Etc.
Tools to include simulating portfolio switching based on various criteria would really help. But having hundreds of OV clients running similar simulations (trying to find the answers to the same questions) puts a hardship not only on the clients, but also on the OV systems.
Why not have the engineers figure out what works? Why not put a bow on the best 6 - 12 "configurations" and make it easy for users to deploy? And why not engineer evolution into the core of OV so the engineers don't have to constantly tune the configurations?
These ideas have been floating around Nirvana for years. ARM, DSS, strategy voting, etc. They just seem to get bogged down somewhere between the white board and profitable deployment.
I think Nirvana should offer cash rewards for proven ideas, then integrate them into the core.
Kudos on your hard work and initial findings. I hope Nirvana sees the value and devotes some quality engineering to 1) proving the concept and then 2) implementing an elegant solution.
Cheers
Keith
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Thanks Steve(2)...
The returns in these charts are simply from closing down trades and starting new with the next month's portfolio. Since a switch only occurs around 45-50% of the time, the current portfolio often doesn't need to be closed - and the math is simpler. We have a SS that will accurately calculate the results allowing trades to stay open from the previous month and then be closed at their normal time via OV while adjusting the new trades according to the 200% limit. It's quite helpful, but doing that is tedious (downloading 185 months of trades). What I've found when I've done it is that the returns are often better than these because of the value of letting the trades close.
I haven't noticed any appreciable difference in TPM or average % Invested.
At this point, our process looks at previous characteristics of the 2 portfolios and then chooses one or the other to trade the next month (no peeking ahead...).
From 1/1/07 to 1/1/14 (the posted charts) the portfolio chosen for the following month has proven to be the best choice of the two 70% of the time.
Keith:
You're right about the process being easier if automated. Some of our macros run 8 to 10 hours in excel. (Steve's an amazing SS design and statistics guru - I'm just the macro designer and "grunt work" guy)
My hope is that Nirvana may be providing some new tools in the near future that will help Steve & I - and everyone here - test our theories in a much more efficient manner. I'm not planning on any major new "research" until I see what those tools are.
Mark
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
According to a Feb 2010 article in the Morningstar Advisor, between January 1926 and April 2009, the S&P-500 had a monthly mean return of 0.91% and a monthly standard deviation ("sigma") of 5.55%. If the market follows a Gaussian or “normal” distribution, there should only be a 0.1% probability of the index exceeding a 3-sigma event. In other words, the index should only have dropped by more than 15.74% (0.91 - [5.55 x 3 standard deviations]) once between these dates. Yet over those 83 years, there have been 10 losses of three sigma or higher in that market index. That equates to about a 1% probability of a 3-sigma event. Thus, it seems that the S&P-500 is 10x risker than we all thought because of its fatter tails (the parts of the bell curve approaching zero).
 
But the fact remains that a 3-sigma event is still very rare in stock returns. So, three-sigma was the high goal that Mark and I set for ourselves. We want to be 99% confident that our switching algorithm is better than random chance. Can we do it?
The attached graph shows a 1000-iteration Monte Carlo experiment where we randomly chose between our two source portfolios. The error bars are three standard deviations above and below the mean Monte Carlo equity (dotted line). The red line is the return from our algorithm-switched portfolio. Note that the return on the switched portfolio is showing a better than three-sigma improvement over random chance for most of the 7-year period!! (P-value=0.01 using one-tailed paired t-test for means of Switched vs. Monte Carlo).
Yes! We can now say our portfolio switching approach is statistically valid,. And over the last 7 years, as simulated in OV, it generates a 20% higher average monthly return (5.97 vs. 5.0) with a 1% reduction (8.8 vs. 8.9) in risk (standard deviation) compared to our selected benchmark (Nirvana Club’s ARM-4 Margin). In the accompanying distributions, notice how the return has been shifted to the right.
Granted, we don’t have 83 years of data to compare against the S&P500, but both of these portfolios outperform the market, by 2.8 and 2.2 points respectively on a risk-adjusted basis. Ed and the Nirvana staff deserve all the kudos for that one!
Attached file : SD Chart.jpg (103KB - 752 downloads)
Attached file : PastedGraphic-15.png (50KB - 751 downloads)
Attached file : Monte Carlo To 3 Sigma.jpg (116KB - 768 downloads)
Attached file : PastedGraphic-14.png (74KB - 731 downloads)
|
|
kmcintyre
Posts: 890
Joined: 10/11/2012
Location: Portland, OR
User Profile |
Nice work, Steve!
Do you think results from testing two portfolios is sufficient to draw the conclusion that portfolio switching will improve results on most sets of portfolios? In most markets?
Intuitively, portfolio switching makes sense provided the frequency of sampling the market is substantially higher than the frequency at which the market charges "personality". Corollary the Nyquist theorem.
If the market changes personality faster than the portfolio switching algorithm can deal with, the results will be whipsaws leading to potentially worse performance than holding any single portfolio. The worst case is probably having the market change at the same frequency as the portfolio switcher samples.
I'm not saying the sample frequency has to be 2x faster than the shortest market personality change, as Nyquist would imply. I don't believe the markets are sufficiently cyclical to firmly apply Nyquist. Even with Nyquist one has to pick a filter frequency above which cycles are simply ignored.
So intuitively portfolio switching can be subject to the same whipsaws seen with moving average crossovers, or any other technical indicator. Pick the wrong sample frequency and I assert it can be harmful.
Is monthly switching right for EOD traders? I like it. Probably right for monitoring quarterly changes. A good place to start...
I ^REALLY^ applaud you bring statistical analysis to the table. I'm totally convinced that there is enough empirical evidence to warrant further investigation. I hope you get the tools required to make that task much less laborious.
I think you too should be a research scientist for Nirvana! :-)
Cheers!
Keith
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
>> Do you think results from testing two portfolios is sufficient to draw the conclusion that portfolio switching will improve results on most sets of portfolios in most markets? <<
I started my career in clinical research as part of the team that developed, studied and obtained regulatory approval for ibuprofen. Now, after 30 years, can I tell you that ibuprofen is absolutely safe and effective under every conceivable scenario? Absolutely not. But, I can say with a level of confidence that, based on well-controlled studies, it is a pretty good choice to treat a headache. And, with my limited understanding of how ibuprofen works at the cellar level, I can say it will probably work just as well for other types of pain. If I’m having a heart attack and have left my bottle of aspirin at home, I will probably take it for that too, but I have no evidence to know if it will work.
Give me 30 years and few $billion and I still won’t be able to answer your (obviously rhetorical) question. All Mark and I have said is that the result we are getting with our particular system has some statistical validity, and it looks like the technique has fairly broad potential. For example, the accompanying attachment shows our switching technique applied to another set of portfolios.
>> IMHO, there is no benefit to sudden changeover. A blended shift makes more sense to me. <<
Linus Pauling told us that vitamin C could cure the common cold. After more than 30 trials over 40 years have shown otherwise, a lot of people still swear it works. As traders, and as patients/consumers, we all have our entrenched beliefs, most of which have little empirical evidence to support them.
I can test a hard-cut switch each month and reasonably quantify the sources of error (commissions, MOC vs. MOO entry, the robustness of the underlying OV simulations). Conversely, I can’t currently simulate a phased-in switchover with enough confidence to draw any conclusion because it introduces a new hard-to-quantify source of error (which trades will be missed while the equity is tied-up by the carry-over trades). It’s a completely different scenario and the outcome of our work should not be extrapolated to that new scenario
>> Why not have the [Nirvana] engineers figure out what works…put a bow on the best…and make it easy for users to deploy? <<
Just like we all want pharmaceutical companies to give us a little pill that instantly cures our cold with absolutely no side-effects and demand they do it immediately, we want companies like Nirvana to give us that elegant solution that guarantees us huge profits. Oh, if it were only that easy!
Mark and I put in hundreds of hours testing various approaches, most of which were failures. And there are thousands more (as many posts have point out) that we haven’t tested. For the same reasons we shouldn’t expect the pharmaceutical industry to develop a new drug until there is enough evidence to support that a viable treatment is within reach, it is unrealistic to expect Nirvana to undertake low-level experimentation until there is a clear market demand and associated commercial viability. In medicine, basic science is done by academia using government funding. For trading systems, that basic science can only be done by interested zealots like Mark and I (and you guys that reads these posts). Let’s recognize that the handful of engineers at Nirvana, as intelligent and creative as they are, could never accomplish as much as several hundred N-Club members all approaching the difficult basic research tasks from different angles. Let’s all rise to that challenge, test our theories and make our proposals, and let Nirvana focus on those with the broadest commercial viability and the greatest benefit to all.
Attached file : PastedGraphic-16.png (73KB - 703 downloads)
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
| Just like we all want pharmaceutical companies to give us a little pill that instantly cures our cold with absolutely no side-effects and demand they do it immediately, we want companies like Nirvana to give us that elegant solution that guarantees us huge profits. Oh, if it were only that easy! |
WELL put. That's the heart of the matter.
I can personally attest that Mark and Steve are at least as qualified as scientists and technicians as the good folks at Nirvana are. I saw some early versions of their work ... and I'm not too much of a slouch with Excel or math ... and it really really impressed me.
This problem has so very many variables, it is what thermodynamicists (my field) describe as a "virtual open system". That is, there are too many independent variables to manage properly to do a nice clean single-goal goal-seeking analysis.
There are many different solution methods. Mark and Steve took several months, as I recall, fiddling with different approaches, till they finally settled on a path that only diverged a couple more times. What they have now, I'm not privy to ... I'm sure it's certainly very very robust.
But, having said that ... it's just ONE possible solution-sieve. Who's to say it, or any other method that another team might come up with, is "the best"?
I'm very glad that Nirvana is keeping things open ended, rather than totally black-boxing THEIR particular "solution" (as if there were such a beastie).
The plans that Ed discussed in the webinar is to offer a Pro level that allows ANYONE with coding experience the ability to develop their own "sieve" that goes far beyond OScript formulae. M&S will likely be able to port their stuff into that engine. Keith will be able to program his own approach there. And others will too, I'm sure ;~)
At the same time, N is planning a very SIMPLE "click one button" approach that allows a tyro user to pick one of many Nirvana-optimized "motif" solutions.
Those, in addition to the Member level with the V2 features plus more, that we have today.
This all was discussed in the webinar, so I hope I'm not speaking out of turn. It sounds like a GREAT plan to me!
|
|
Mark Holstius
Posts: 744
Joined: 10/11/2012
Location: Sleepy Hollow, IL
User Profile |
Since I mentioned it in your webinar this evening, here's the run from the latest version of the algorithm... 3X the equity of your excellent ARM4, Ed.
The tools you're developing should be a great help to everyone!
Mark
Attached file : Latest run.jpg (151KB - 623 downloads)
|
|
Bruce Britt
Posts: 81
Joined: 10/11/2012
Location: Louisiana
User Profile |
Mark and Steve,
My hat goes off to you guys for your exhaustive research and effort - and especially for sharing you work in an effort to accelerate the development process for the benefit of the group as a whole.
Kudos also to Ed and the N development team for recognizing the significance of your findings and for taking immediate action to provide the tools required to implement you concepts.
I’ve only been a N Club member for 2 years, but at this point, I’m convinced that joining this group is one of the best decisions I have ever made.
Thanks, -Bruce
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
Hear, hear!
Hip hip hooray!
|
|
JayP
Posts: 6
Joined: 10/11/2012
Location: pomona, New York
User Profile |
Mark and Steve
Ed just mentioned on the webinar that I had an interesting, but too complex question and suggested I post it on the forum. This seems like the best spot.
When "you" set up the choices for optimizing in the PRO version, have you considered working from their "Level" of membership.
It would be a different set of strategies for those who can utilize only 5 total systems (bronze) versus the other levels 10(silver) and 20(gold). Are we considering making the "Pro" level available to all, or just those at Gold and Platinum levels.
Getting meaningfully into Portfoio Wizard might only be possible for those of us who are Platinum members. Would there be a way of "suggesting" the optimal number of strategies if one were a platinum member. I'm thinking of Mark's thread on optimal number of strategies and how it might tie in .
I love what you guys are doing. Thank you so much.
|
|
Steve Mayo
Posts: 414
Joined: 10/11/2012
Location: Austin, TX
User Profile |
Hi Jay,
I can't speak to Nirvana's business strategy, but from what we've been discussing the port switcher/balancer tool will work with whatever portfolios you have, regardless of your membership level.
If you have, say, Bronze level, you are limited to portfolios that have only 5 strategies. You can still assemble lots of different portfolios, and the OV-Pro swticher will use whatever ports you have, hopefully resulting in a "switched" portfolio/system that does better than any of those.
Conceivably, and I'm only postulating now, OV3 and the Elite Trader program will enable a Bronze or Silver user to subscribe to larger portfolios (more strats) than he/she could otherwise build at his/her membership level. For example, Mark or I might publish a "system" that switches between 2 or 3 of our best portfolios. Under the Elite program, I would think any OV user could subscribe to it regardless of membership level, but to be fair to Gold members, I would expect bronze/silver users might have to pay a higher monthly fee.
That said, the cost to upgrade to Gold membership is trivial compared to the dramatic improvement you get from using 10 or more strategies in a portfolio. With a bit of work in Strat Lab and with the right list, one can now get higher returns with only a few strategies, but it generally takes more risk compared to the smoother ports you get with 10 or more strategies.
|
|
Jim Dean
Posts: 1059
Joined: 10/11/2012
Location: L'ville, GA
User Profile |
Also, remember that the various level limits refer to how many strats are IN USE at a given time. So, if there are 100 different systems avail, then the number of 5-Strat-at-once portfolio permutations allowed for bronze level is 100*99*98*97*96 (you do the math) … and that totally disregards the fact that each of those 100 systems can be combined with dozens of different lists and an effectively infinite number of conditions - thus the number of portfolio variations are literally endless even for a Bronze subscriber. The only limitation is that any given portfolio may have only five strats active at once. But if the switchover is made monthly or whatever from one portfolio to another, then in any given year, potentially 12*5 = 60 entirely different systems/strategies might be called upon.
Truly a wonderfully versatile architecture, offering unparalleled degrees of freedom for minuscule amortized cost.
|