Market Timing - only 4 trades since 2000

Here is a model that hardly ever trades, but still shows an annualized return of over 13% since 2000, and a low D/D. It switches between USMV and IEF.

I am using a recession signal uploaded into a series together with my Portsum timer.

I guess this model will not be popular at P123, because it is all about not trading.


I would be. that is a dream… !

I am using the iM-Short Leading index uploaded into a series called “iM-SLI” together with a multi-model timer which uses a number of independent models called Portsum.

Buy: Eval(close(0,GetSeries(“iM-SLI”))>=0 & $portsum>5 ,TICKER(“USMV”),TICKER(“IEF”))
Sell: Eval(close(0,GetSeries(“iM-LLI”))<0 ,TICKER(“USMV”),TICKER(“ief”))

The rules are self-explanatory. So simple, just get out of stocks before recessions.

From Figure-2 it is evident that we are not near a recession. So happy trading equities!!


That’s impressive!
I think this new upload feature is going to be a game changer!

I think an advanced P123 workshop/ conference will be really useful. The idea sharing, networking, etc., can really improve us all. I also want to take a moment to Georg. Yuval, Marc, Andreas and the many people over the years that have selflessly shared to help improve our understanding and utilization of this fantastic product.

Not sure how that works out, since IEF has an inception date of 2002-07-22 and USMV has an inception date of 2011-10-18.

Randy, you must be new to ETFs. P123 has synthetic data for numerous ETFs prior to their inception dates.

Here is another USMV-IEF model which uses ECRI’s WLI and Portsum. Only 16 trades and 16% AR. Slippage= 0.1% is included.
The model never under-performs USMV.


Georg,

you show market timing with hindsight. As you may know from your own market timing models, OOS performance looks different…
https://www.portfolio123.com/app/r2g/summary?id=1098426

Florian, the model you a referencing has totally different rules and ETFs than the USMV-IEF model.

The Lazy Person’s Market Timer - 4 Trades since 2000 has only 4 trades because there were only 2 recessions since 2000. This model only does better than the benchmark because it is out of USMV during recession period, otherwise it obviously matches USMV.

Any reasonably good recession indicator can be used. One does not have to use the iMarketSignals’ indicators. I recommend the publicly available Conference Board LEI (CB-LEI) - you have to buy the series G0M910 for $250 and then calculate the smoothed 6 month annualized growth rate from it (CBg).
https://www.conference-board.org/data/bcicountry.cfm?cid=1

Here is the formula CBg:

“MA1” = 4 week moving average of the CB-LEI
“MA2” = moving average of MA1 over the preceding 52 weeks
“n”= 52/26.5
“m”= 100
CBg = [m*(MA1/MA2)^n] - m

Her is a model with only 5 trades which uses the Conference Boards LEI Growth with 2.35 added to it to make zero the recession warning line. Annualized return is 12.3% and max D/D= -14. That is a nice safe investment model which you don’t have to worry about.

Here are the rules:

Buy: Eval(close(0,GetSeries(“LEIg+2.35”))>0 & sma(50,0,$SP500)>sma(200,0,$SP500),TICKER(“USMV”),TICKER(“IEF”))
Sell: Eval(close(0,GetSeries(“LEIg+2.35”))<-4 ,TICKER(“USMV”),TICKER(“IEF”))

Everybody knows the SP500 50/200 day MA crossover system(MAC). Problem with it that you get whipsawed often and it is not very profitable on its own.
In this model here you get into USMV when the SMA50 moves above the SMA200, and only get out of USMV when the recession indicator moves below -4, not when the MAC system signals it.

EDIT: The model returns a bit more (12.5%) with this buy rule:
Eval(sma(50,0,$SP500)>sma(200,0,$SP500),TICKER(“usmv”),TICKER(“IEF”))


Georg,

Your having brushed aside Florian’s comment, simply pointing out that the system to which he linked was different from the ones your showing now, is completely unacceptable and, whether you realize it or not (probably not), subjects the portfolio123 community, and by way of association, Portfolio123 as well, to ridicule (forum posts are not the only opinions expressed so don’t assume you truly know how you are perceived).

Florian is pointing out as politely as possible, that for all the backward-looking super-charts you post, you have shown no ability to actually deliver on anything when real money is invested. DMs have been around for seven years now and your track record is an embarrassment, which makes your recent amateur-hour cheap shot against the recent struggles of my Cherrypicking the Blue Chips model (which unlike anything you ever came out with, delivered a lot over a fair amount of time) all the more bizarre.

Portfolio123 delivers a lot of great tools, but tools misused can lead to horrible things (give a chain saw to a five-year old to see what I mean — oops, another kiddie table-type metaphor). If you want to build credibility, you really need to cease and desist from using 1/2/2000 as start dates in charts you show on the forum (even though p123 let’s you do it). Limit your publicly-shared backtests to one year and be humble — unless and until you can start to show some measure of out-of-sample success.

I strongly suggest that you remove all of the charts you posted in this thread. Talk instead about your ideas, why you think they’ll work, and invite the p123 community to follow along in real time as you publicly trace forward progress (post more DMs, and rather than hiding from their out-of-sample results, evaluate and discuss them, and learn from them as you go).

But whatever you ultimately decide to do or not do, please stop assuming you are generating favorable impressions by acting as if your demonstrated ability to predict the past has any meaning besides a cute game.

Sorry to be so blunt, and I know posts like this don’t thrill Marco who would prefer to see a more collegial tone on the forum, as would we all. The reality, however, is that quants have been getting raked over the coals harshly and frequently in recent years and it’s been getting worse, not better. (The physics envy article Jim found, at my prompting, was just an early shot across the bow.) To try as best I can to preserve Portfolio123’s reputation and branding, it really is important, albeit regrettable and unpleasant, to make it clear that what you are doing in this thread, and others like it, is not at all representative of good use of the powerful and extremely beneficial tools Portfolio123 works hard to provide.

If you like public discussion of results, focus instead on what you discuss re: riding coat tails on ETFs. That’s actually a good idea and seeing another post in that thread from somebody who took 8 hours to create a csv of holdings to upload (ouch!!!) prompted me to communicate to Marco and Yuval off line about the need to really get serious about licensing ETF constituents. That sort of task should be do-able in 8 seconds (or less), not 8 hours. But working in this way as best you can with the capabilities at hand right now (you can create constituent files much more quickly if you stop feeling compelled to max out on backtest length) would make for much more interesting, valuable and credible forum discussion — but even here, exercise self-discipline and aim for what’s likely to be real, not super charts. At least this, unlike your business-cycle work, can give bona fide o-o-s results without waiting for years to pass.

Marc, I did not want to rub in the point that this model you and Florian references did actually quite well. As a matter of interest it has shown a 5 year return of 122.89% versus SPY of 78.98% and was one of the very few designer model which has outperformed SPY over the last 5 years. It has done better than your model Cherrypicking the Blue Chips - Standard which only produced 63.26% over this period.

I am confident that the recently revised model will continue to out-perform SPY.

Typically a market timing model cannot show excess returns over SPY in up-market conditions. It out-performs only by being out of stocks during down markets. So let’s see what happens when the next recession is upon us.

Below is a screenshot of Best(SPY-SH) Gains for Up & Down Markets showing the 5 year performance.


Marc, if you search the forum you will find that it was actually my idea, more than 5 years ago, to ride the coat tails on ETFs. I am glad that you have recommended to the P123 Team to get serious about licensing ETF constituents and I am looking forward to this feature. I have two models live, using VDIGX and USMV, which show the benefit of doing this for more than 5 years.

As far as the business cycle is concerned, and how one can benefit from knowing when a recession is imminent, you don’t have to use my indicators. By implication, you also belittle the work of the Conference Board (US Leading Indicators), the people at the Chicago Fed (Recession Forecasting With The Federal Reserve Bank Of Chicago's Newly Released Brave-Butters-Kelley Indexes | Seeking Alpha), and others who produce those business cycle models. That’s just not fair.

For interest you may want to have a look at my recent Seeking Alpha posting and the comments to “Robust Recession Forecasting With A New Long Leading Index For The U.S. Economy”

It may be just nonsense to you, but others find it quite interesting.

I agree with Marco. But let’s face it, it was you who saw fit to respond to a perfectly good suggestion, namely to put some artificial noise into a backtest, that the person (Engineer) proposing it should study your manual first in order to sit at the “adult” table. I found that un-collegial and he found it insulting.

Marc, we are in the USA not in Russia.
You should know that one cannot remove a chart that has been posted. Needless to say, even if I could, I would not remove those charts because you don’t like them. Perhaps other people find them interesting.

Hi Georg,

The reason why I commented in this thread to begin with was the following conversation:

You wrote [quote]
I guess this model will not be popular at P123, because it is all about not trading.
[/quote] and RTNL replied [quote]
I would be. that is a dream… !
[/quote]

So what I disliked is the fact that other members in the forum get excited about backtests which are likely not reproducible in the future. Whether you use one ETF or the other or one timer or the other doesn’t matter: it is all done with hindsight knowledge of past recessions - while future recessions might manifest themselves very differently.

I really appreciate all your inputs over the years, and I will equally do so in the future. However, I would also appreciate if you take on board the learnings from OOS performance.

Now coming back to your Best(SPY-SH) Gains for Up & Down Markets designer model you advertised this in September 2013 as follows [quote]
Using a web-based trading simulation platform, our ranking system, and specific buy and sell rules, this model would have produced an average annual return of about 29.3% from January 2000 to end of August 2013, versus 2.6% for a buy-and-hold investment of SPY over the same period.
[/quote].
Fast-forward over 6 years and three model revisions later a potential subscriber would have spent 75 x 45$ = 3375$ for less than 5% outperformance over the SPY (see below).

Hi RTNL,

what do you think? Does this still sound like a dream?


Hi Florian,
Thank you for your measured and professional posting. I respect your views, you have a good point here. Why would anybody want to subscribe to such a model that hardly keeps up with SPY. You can ask the same question about all the other DM’s that needlessly trade stocks and underperform the index.

The models based on recession timing show that it is not necessary to aimlessly buy and sell stocks to make an acceptable return in the market. Of course, if you can trade stocks profitably with the P123 tools you can do much better. Some of your models do this.

So I do accept your critique, but as I said before, market timed ETF models can only work if we have some major prolonged down-markets. That has not been the case for the last 10 years. Wait a little longer, the recession is not too far away, then my market timing model will perform better. In the meantime if they track SPY they are doing o.k.

Thanks.

Marc,

Marc I agree with most or all of what you say.

So maybe we are supposed to do it like this?

I like Yuval’s ideas here but machine learning—or manual optimization like the above–halfway done, it is exactly like what Georg is doing.

You have said you think Yuval should be free to express his opinion, I agree. As long as that is your position then you should not single-out Georg.

If you worry about P123’s reputation then P123 should stop presenting inconsistent views on the forum and singling out members.

And maybe hire a consultant or have someone there take a course or open a textbook about some of the ideas Yuval expresses here and expand upon them.

But look inward first.

P123 could consider embracing Yuval’s ideas while working to learn the best practice for avoiding some of the problems that you criticize when criticizing Georg. You should be talking to Yuval about this. And both of you should be working with Marco.

[b]Marco can expand upon Yuval’s ideas and control overfitting at the same time. Marco has hired an AI expert who–when it was all done–recognized that the methods he used gave predictions that were not better than random.

People can use Yuval’s ideas or ones like them–and avoid overfitting at the same time. Marco know this.[/b]

Marco knows how to do this and/or whom to consult with.

BTW, even if you agree with everything Yuval does in the above quote it takes more than 15 seconds.

-Jim

As with any model historical returns are just that. I see no reason to hide a display of what happened in the past. All model’s come with a disclaimer about the expectation of future returns and I haven’t seen anyone state that they can expect X% going forward. Again this brings up my point about not judging a designer based on past model successes or failures. Just like it’s not fair for me to say all Chevy cars are terrible because they made a few bad cars in the past. The next car they release may be the best car of the decade. I think most of us learn from our mistakes and we continue to get better.

I for one thank Georg for all his contributions. I think he is one of the more helpful contributors on P123. Not because he shows amazing backtests nor because he seems to have mastered historical market timing (as stated OOS results pending but we can’t do anything about that, we don’t have a fast fwd button). Georg has shown that he is very good at P123 syntax and I have learned a lot of clever tricks from his gracious gift of sharing. I don’t think I would have figured most of them out from P123 documentation alone. There is a LOT of flexibility within the syntax of P123 and unfortunately most people would not be able to exploit that from reading basic descriptions. I’m not sure if anyone would be able to write an advanced user guide and capture all the possibilities (that would be a really long guide).

On the topic of thanks, I would like to thank all the members that have contributed syntax tricks (and market insight). You make the tools all the more powerful to the rest of us. P123 is a language all of it’s own and like any language you can never remember every word in the ever expanding dictionary. There’s usually a word or words to express what you want to say but sometimes we need help finding the right words.