Combining Vol, Economic, Trend, and Momentum Strategies

A place to talk about speculative investing ideas for the optional Variable Portfolio

Moderator: Global Moderator

pmward
Executive Member
Executive Member
Posts: 1731
Joined: Thu Jan 24, 2019 4:39 pm

Combining Vol, Economic, Trend, and Momentum Strategies

Post by pmward » Wed Nov 25, 2020 3:30 pm

Ok so I'm going to say up front that this is going to be quite the brain dump, but those quant nerds out there will enjoy it. Kbg requested I share some info into what I am using for my quant strategy these days. Instead of giving the exact custom recipe, I'm going to present the base starter recipe that others can start from. I'll also include some open source and closed source options where possible.

So first off from a high level what is the point of this portfolio? What problem is it trying to solve? Well, as we all know most quant strategies do great 90% of the time, but there is the 10% of the time that they fall flat on their face. So what we are trying to accomplish is mix a bunch of diverse strategies together to make one super robust whole that is greater than the sum of its parts. The combined strategy performs way better than any of the 4 strategies in isolation. We want to be invested aggressively into the areas of the market with momentum (preferably with leverage) when the skies are blue. We want to be in some form of safe asset in any period of abnormally high near to medium term volatility, or any period of combined economic weakness AND negative price trend. We also want the risk off assets to account for possible risk off periods where interest rates are going up. So basically TLDR is we want to be super choosey on when we are in the market, and when we are in, we want to bring out the big guns. When we are out of the market we want to perform well regardless of whether interest rates are trending up or down. Quite a big task at hand. So let's dig in.

So for the impatient I will start off with the closed source all-in-one recommendation of Paul Novell's VolComp Global TAA. Then from there I will break apart each piece and share any open source alternatives and my understanding of each piece so people can create their own systems. So you can subscribe to his QuantPulse service to get this VolComp Global TAA on his website https://investingforaliving.us

No Leverage:
1999 - July 2020: 15.88% CAGR, -21.14% max daily drawdown (SPY 6.49% CAGR, 55.19% max daily drawdown)
2009 - July 2020: 16.20% CAGR, -17.38% max daily drawdown (SPY 13.92% CAGR, -33.72% max daily drawdown).

Apr 2 2008 - Nov 20, 2020
No leverage: 18.49% CAGR, -17.95% max daily DD
2x leverage in "risk-on" periods only: 29.38% CAGR, -31.85% max daily DD
SPY 10.07% CAGR, -51.49% max daily DD

So all in all not too shabby. It both out performs in bear and bull markets. It also performs better with leverage than without. It doesn't beat SPY every year, but it never significantly underperforms, and it never gets its face ripped off.

Now there is plenty of room in this framework for customization to spice things up. Warning, only continue on if you have your quant nerd hat on. So there is really a combination of 4 quant strategy layers here. Volatility, economic data, trend following, and relative momentum. I will dig into each layer individually in a way that respects Paul's closed source format. My hope is to share some open source alternatives where they exist, and leave the group discussion to come up with their own quant algorithms for these pieces. Also, I'm curious to see if kbg or anyone else can come up with any improvements I haven't considered. Even though I created my own recipe based on this starter recipe I did still sub to Paul's service as a way to support him, as not only would I not be aware of this type of strategy if not for his public blog (not to mention a shout out to Ocho for bringing the VolComp strategy to my attention last winter when I was looking at different volatility quant strategies), but also Paul was gracious enough to answer quite a few of my emails when I was first going down the rabbit hole. So without further ado.

1) Economic data and trend following. I don't really need to get too deep into this one, as it's already been discussed on this board via Novell's SpyComp. Basically, the idea is to look at economic data that has historically been a good predictor of recession. When any of that data is red, then you revert to some form of trend following (like a moving average) to tell you if you are "risk on" or "risk off". Some info on Paul's closed source SpyComp: https://allocatesmartly.com/paul-novell ... -spy-comp/. For a more basic open source version you can use something like this here: https://allocatesmartly.com/philosophic ... ing-redux/.

2) Volatility. Basically the idea is looking at the VIX futures curve to tell us when it is a normal or abnormal volatility environment. In a normal environment the VIX curve is in contango. This makes sense, because 6 months from now is more uncertain than today... so the 6 month VIX contract should be priced higher than the three month contract, and likewise the 3 month contract should be priced higher than the front month contract. This is also why volatility has a "cost of carry" associated with it. If you buy these contracts and roll them in all "normal" times it will cost you money to do so. So what does it mean when the VIX futures curve flips into backwardation? When the more known front months get priced higher than the more unknown longer dated months? It generally means very bad things. So we want to avoid investing during these periods of abnormal volatility. Now this is closed source, so I will just say that it's not as simple as "green" when in contango and "red" when in backwardation. Paul does use some other common quant strategy tactics to smooth and manipulate the data, especially on the buy back. You can dig into the following reading material, he lists all past trades for the model going back to 2008. Based on what he has provided if you feel like it you can try to reverse engineer what he did, or at least build something on your own in the same spirit that performs well. I am not aware of an open source strategy that is similar unfortunately. Read all the following in order:

https://investingforaliving.us/2019/12/ ... rve-model/
https://investingforaliving.us/2020/03/ ... rve-model/
https://investingforaliving.us/2020/06/ ... rformance/
https://investingforaliving.us/2020/07/ ... rve-model/
https://investingforaliving.us/2020/07/ ... rve-model/
https://investingforaliving.us/2020/08/ ... ve-models/

3) So we have these two separate defensive strategies. How do we combine them into one? Quite simple. When either #1 or #2 is red you are in the "risk-off" asset (always no leverage). When both #1 and #2 are green, you are in stocks (optionally dial in leverage how you like). This fits the bill from above about being super picky on when we are in the market. We have economic data, trend following, and volatility telling us when it is ok or not to be in the market. Worth noting, the economic data and trend trade at end of month only, the volatility can trade any day. Futures markets close at 4:15 (later than stock market) so you have to wait for the close. So any volatility curve trades are always on the day following the signal (trading at the following close has a slight edge over the open).

4) Momentum. This is another one I don't have to give too much info on as it has been thoroughly covered on other parts of this forum. Paul's version uses a dual-momentum of U.S. and foreign stocks. But any momentum strategy will work here. If you choose a higher octane strategy you can get some extra juice out of this. You can do simple binary pairs like dual momentum, or momentum vs value, or you can also do a basket of ETF's and pick the top x ETF's for the next month, or you can even do an individual stock quant momentum strategy. Whatever you feel is going to give you the alpha and diversification you need to meet your goals. For risk off Paul uses dual momentum between TLT and BIL to decide if in cash or TLT. This helps safeguard against periods of weakness that coincide with rising interest rates (for example it went into cash in Q4 2018 and that worked out way better than going into TLT). I've also kicked around the idea of using a 3 way between gold, TLT and cash with either pick top 1 or top 2 (or an intermediary like if GLD is #1 pick top 2, else pick top 1). This is where we can really have some fun adding some custom flavor to the recipe.

So as my portfolio stands today, I am in a customized version of the above in 70% of my portfolio, with 20% static gold, and 10% static cash. For the cash I am keeping it at 10% until I reach a hard number I have in mind, then I will keep it at that hard number and increase the quant strategy from there. For gold... I would love to find a way to incorporate the gold into some form of a tactical strategy (either combined above or in a different strategy altogether) so I am not holding the bag during the bear phases. But for now I have found no TAA strategy I love using it. In the HB model I would be in stocks in "prosperity" I would be in bonds during a "deflation", and in "inflation" or "tight money" I would be in cash (and I have the side car allocation to gold).

One last thing I should mention is the biggest con of this strategy... taxes. You're looking at a 3-4 trade per year average strategy if you follow the vanilla Novell offering. It's quite rare that you make any long term cap gains. However, this is one of those strategies that is high octane enough to make it worth while. Like if my job offers me a raise I'm not going to tell them no because I don't want to pay extra taxes. If I can out perform SPY by double digit CAGR... it's well worth paying a good chunk of that alpha in taxes. I've done the math, even assuming all short term cap gains both the unlevered and levered version still smoke buy and hold forever SPY. And that doesn't even account for 401k, IRA, HSA, etc that are tax sheltered.
D1984
Executive Member
Executive Member
Posts: 460
Joined: Tue Aug 16, 2011 7:23 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by D1984 » Thu Nov 26, 2020 2:22 am

Those returns look rather mouth-watering but a few quick questions spring to mind:

One, while I am aware that the economic data based SPY-COMP model has been backtested to the mid-1960s none of the rest of the models linked (incl the VIX vol curve based one) seem to have been backtested back much beyond 2008. VIX futures have been around since mid-2004 (and Barclays, S&P, and several others have successfully simulated them back to the early 1990s or further), the VIX itself (or VXO which is very similar) has been around since 1-1-1986 and could easily be backtested at least a few years before that because the VIX index itself is based on option prices and S&P 500 and S&P 100 index options have been around since early 1983. Given all of the above, why does the VIX volatility curve based strategy only go back to mid-2008? My concern here is that during the mid-2008 to present period we have been in a rather unique environment (low rates that are generally falling over time to boot...falling rates act like rocket fuel for risk asset prices, a Fed that is ready and willing to intervene every time the market so much as hiccups or sneezes, people having to buy equities because the returns on the alternative--fixed income--is so pitiful, etc) and there has been no "out of sample" backtesting so to speak for any other kind of environment. Even if you go back to 1999 we have likely still more or less always been in a market environment that always had the "Greenspan Put" (and then the Bernanke Put, and the Yellen Put, and now the Powell Put, etc); you'd have to backtest until the better part of a decade before the 1997/1998 market turbulence caused by the Asian Financial Crisis, LTCM failure, and Russian defaults in order to really see how a strategy would perform over a reasonably long period of time without a Fed always willing to shovel out the liquidity every time the market got spooked a little bit.

Two, is any strategy that can deliver 29% CAGR results sustainable over the long-run? If the market returns on average say, 9% a year and this strategy returns 28 or 29% then wouldn't such a strategy eventually grow so large as to be unusable (because you can't time the market when you have grown so big that you ARE the market)?

Three, I just want to be sure I am getting this and am not being stupid and missing something obvious so let me see if I have it straight: Only go into stocks (or some leveraged version thereof) when both the vol curve and econ/trend data are flashing "ON" signals indicating it is likely to be safe. Having decided to do that (go from "risk-off" assets into equities" use dual momentum to choose whether to be in US versus foreign stocks (or whether momentum like MTUM or value like RPV has done better over a given time period, or whether QQQ or some kind of SCV have done better over a given time period, etc or something similar to decide what type of stocks to be in, and then use that to determine which category of stocks you wish to be in). When either the econ/trend or the vol curve (either one or both) flash a "danger" signal, go to your safe "risk-off" asset; when it comes to what to choose as the safe risk-off asset use dual momentum to choose what kind of safe asset to be in (presumably either cash vs LTTs, cash vs intermediate TIPS, cash vs gold, bonds vs gold vs cash, etc). Do I have this right?

How much time has the model spent in risk-on assets (presumably stocks) and how much time has it spent in risk-off assets?

Fourth and finally, if I subscribe to Mr. Novell's service does he have anything on his site that can run a quick backtest if given data in either CSV or XLS format for monthly (or if need be, daily) returns? What if I wanted to simulate, say being in either TQQQ or TNA as the risk-on asset (using dual momentum to decide which to be in) and then simply staying in STTs (like VFISX...or a simulated blend of cash and FIGTX that is a dead-on proxy for VFISX that would allow the backtest to be extended to the mid-1980s) during periods when being in risk off assets was indicated...could I do that (because I have backtested data for TQQQ to the early 1980s if needed)?
pmward
Executive Member
Executive Member
Posts: 1731
Joined: Thu Jan 24, 2019 4:39 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by pmward » Thu Nov 26, 2020 7:55 am

D1984 wrote:
Thu Nov 26, 2020 2:22 am
Those returns look rather mouth-watering but a few quick questions spring to mind:

One, while I am aware that the economic data based SPY-COMP model has been backtested to the mid-1960s none of the rest of the models linked (incl the VIX vol curve based one) seem to have been backtested back much beyond 2008. VIX futures have been around since mid-2004 (and Barclays, S&P, and several others have successfully simulated them back to the early 1990s or further), the VIX itself (or VXO which is very similar) has been around since 1-1-1986 and could easily be backtested at least a few years before that because the VIX index itself is based on option prices and S&P 500 and S&P 100 index options have been around since early 1983. Given all of the above, why does the VIX volatility curve based strategy only go back to mid-2008? My concern here is that during the mid-2008 to present period we have been in a rather unique environment (low rates that are generally falling over time to boot...falling rates act like rocket fuel for risk asset prices, a Fed that is ready and willing to intervene every time the market so much as hiccups or sneezes, people having to buy equities because the returns on the alternative--fixed income--is so pitiful, etc) and there has been no "out of sample" backtesting so to speak for any other kind of environment. Even if you go back to 1999 we have likely still more or less always been in a market environment that always had the "Greenspan Put" (and then the Bernanke Put, and the Yellen Put, and now the Powell Put, etc); you'd have to backtest until the better part of a decade before the 1997/1998 market turbulence caused by the Asian Financial Crisis, LTCM failure, and Russian defaults in order to really see how a strategy would perform over a reasonably long period of time without a Fed always willing to shovel out the liquidity every time the market got spooked a little bit.
You can see linked article here about the extended backtest data https://investingforaliving.us/2020/07/ ... rve-model/ The CBOE doesn't have data prior to 2007 for the 3 month and 6 month contracts needed in the volcurve model. But he purchased data from somebody that calculated the data back to 1993 (he includes the link describing the methods of recreating this past data). It holds up really well going all the way back to 1993. There was a brief period of underperformance in 1996-1998 though. Also, this data is just using SPY for "risk-on" it does not have the momentum overlay, which would improve returns provided you selected a halfway decent momentum strategy for that layer. It would be nice to have further back data, but this is simply the best we can get. I look at the theory of the what the VolCurve is trying to accomplish though... and regardless of increasing or decreasing rates owning volatility has a "cost of carry" so in normal times (ie stock market not crashing) there should always be contango in the VIX curve. I would have to struggle real hard to find a scenario where the VIX curve was in backwardation while the stock market was doing well. It just doesn't make fundamental sense. Nothing is impossible, at least in the short term, but I think the odds of long prolonged periods of VIX curve backwardation with rising stocks would be low. Let's also remember that SpyComp is also there, one would have to fake both out to fake out the whole model. The intent of this model isn't to be right on every single short term move, it's to catch the big secular moves and trends (preferably with leverage) while having protection in place so you don't get your face ripped off in a crash or bear market. We have 2 bull and bear cycles in data. Also in regards to rising interest rates, there is the dual momentum switch between cash and TLT in risk off. TLT is barely above cash right now, it's possible next risk off would be into cash instead of bonds.
D1984 wrote:
Thu Nov 26, 2020 2:22 am
Two, is any strategy that can deliver 29% CAGR results sustainable over the long-run? If the market returns on average say, 9% a year and this strategy returns 28 or 29% then wouldn't such a strategy eventually grow so large as to be unusable (because you can't time the market when you have grown so big that you ARE the market)?
Remember we are using 2x levered ETF's to achieve that 29% return, the unlevered was 18%. And no you would not become that large to become the market, haha. Also, the biggest years of out performance, like most of Novell's strategies are the years that there is market volatility... ie 2018-2020 have been great, 2007-2009 was great. Less volatile years the unlevered portfolio tends to perform inline or slightly below the S&P. Again this also depends on the momentum layer that you implement, I'm just basing this off of dual momentum of VTI and VEU. No system is perfect. This won't beat SPY every year. But I am confident it will beat SPY over any full market cycle. Especially so if you implement leverage, and if you ask me this is a strategy that is just begging for leverage. That all being said, whenever I backtest ANY portfolio I always look at those results with the assumption that the largest drawdown is in the future, and that future CAGR is likely to be lower than past CAGR. So it may not hit 29%.... but even if it performed 33% worse on both metrics... 19% CAGR with a 40% max daily DD is still good enough, right?
D1984 wrote:
Thu Nov 26, 2020 2:22 am
Three, I just want to be sure I am getting this and am not being stupid and missing something obvious so let me see if I have it straight: Only go into stocks (or some leveraged version thereof) when both the vol curve and econ/trend data are flashing "ON" signals indicating it is likely to be safe. Having decided to do that (go from "risk-off" assets into equities" use dual momentum to choose whether to be in US versus foreign stocks (or whether momentum like MTUM or value like RPV has done better over a given time period, or whether QQQ or some kind of SCV have done better over a given time period, etc or something similar to decide what type of stocks to be in, and then use that to determine which category of stocks you wish to be in). When either the econ/trend or the vol curve (either one or both) flash a "danger" signal, go to your safe "risk-off" asset; when it comes to what to choose as the safe risk-off asset use dual momentum to choose what kind of safe asset to be in (presumably either cash vs LTTs, cash vs intermediate TIPS, cash vs gold, bonds vs gold vs cash, etc). Do I have this right?

How much time has the model spent in risk-on assets (presumably stocks) and how much time has it spent in risk-off assets?
Precisely, you got the general algorithm down. Like most of Novell's strategies it does spend much more time in risk on than risk off assets. This one does trade more often though, especially since the VolCurve is a daily signal that could pop any trading day. On an average year the full system listed above using dual momentum has about 4 trades. So on an average year using dual-momentum you would have 1 risk off event and 1-2 momentum trades. This year had 3 risk off periods and 0 momentum trades for the vanilla dual-momentum implementation for a total of 6 trades. It was risk off from Feb 24 - Apr 14, from 4 Sept - 15 Sept, and from 27 Oct - 6 Nov. The first trade was a home run, the second trade was a single (I believe it gained just under 2% on buy and hold SPY in that trade) and the 3rd trade was a fakeout that took a couple percent loss vs buy and hold SPY. YTD stats below for comparison as of Nov 20:

Buy and hold SPY: 12.02% YTD, 33.70% max daily DD

Unlevered VolCurve Global TAA (with momentum overlay) 17.6% YTD, 15.73% max daily DD

*The following with no momentum overlay, just using format of risk-on/risk-off ETF's*
Unlevered SPY/TLT: 15.87% YTD, 15.73% max daily DD
Levered SSO/TLT: 23.68% YTD, 18.47% max daily DD
Super levered TQQQ/TMF: 82.03% YTD, 43.83% max daily DD

So you can see this year alone there are 2 real benefits of adding the VolCurve to SpyComp. First is that it tends to get out of some of the smaller corrections that SpyComp would hold through. Now, some of these like trade 3 this year can be a slight loss. But there are a lot of small 1-3% base hit less than one month long risk off trades throughout the history. These do tend to add up over time, and help one sleep at night when they are using leverage and are sitting out in case a small correction turns into a big one. Second, in the big corrections, like in Feb, it gets out earlier than SpyComp. There was no waiting until the end of the month, this strategy hit the eject button right away.

Now one caveat, if you read all the articles I linked you will see there are two separate VolCurve models, the "OG" model which is what is used in all the data above, as well as the "enhanced" version. The difference between both is simply the buy back. The "OG" model uses some moving averages in a way to ensure that there is a reasonably long time out of market during times Vol spikes stupid high (like this year and 2008). The "enhanced" version does not have this protection, and buys back quicker. That's the main difference, they both have the same sell signal. So this year the "enhanced" did out perform, as it bought back a few weeks earlier than the "OG". But if you go back to 2008 the "enhanced" version had a couple times it got back in the market with a drawdown that the "OG" market was still out of the market for. So for a quick V shaped recovery like this the "enhanced" version will perform better. But in a more standard secular bear market like 2000-2003 or 2008-2009 you will get less whipsaws with the "OG" model. The "OG" model actually bought back after SpyComp this spring. Just some added nuance to think about for those that want to play with different models. All of this stuff is covered in the links I provided in the OP.
D1984 wrote:
Thu Nov 26, 2020 2:22 am
Fourth and finally, if I subscribe to Mr. Novell's service does he have anything on his site that can run a quick backtest if given data in either CSV or XLS format for monthly (or if need be, daily) returns? What if I wanted to simulate, say being in either TQQQ or TNA as the risk-on asset (using dual momentum to decide which to be in) and then simply staying in STTs (like VFISX...or a simulated blend of cash and FIGTX that is a dead-on proxy for VFISX that would allow the backtest to be extended to the mid-1980s) during periods when being in risk off assets was indicated...could I do that (because I have backtested data for TQQQ to the early 1980s if needed)?
I am currently only subscribed to his EconPulse, which doesn't have the VolCurve. I did subscribe to QuantPulse briefly, and when I was subscribed to it I had a lot of email conversations with Paul. I was able to ask him things like you're asking there, and he was able to pop them into the spreadsheet and tell me the result. It is not publicly available though unfortunately. That being said, I am anxiously awaiting Excel to finally release their new stock features out of beta. When those come I do plan to build my own spreadsheet to be able to simulate this kind of stuff. In the meantime, I don't have any tools myself that are able to easily extract the data, other than just looking at the dates of each trade (which are listed in the public blog linked above) and manually calculating each trade on those dates for comparison. What I can tell you is that all the different flavors of relative momentum that Paul includes in his EconPulse that out perform regular old dual-momentum (momentum : value, emerging-markets : equal weight, world factor, etc) all out perform vanilla dual-momentum in this system as well. So really, this momentum layer is probably the piece that is the easiest to customize to tack on a bit more CAGR, as well as to include diversification between different locales and factors. Do keep in mind the number of trades though. The more ETF's you include the more trades you will have to make. You can generally model different momentum strategies on portfoliovisualizer to get a feel for the general relative performance and number of trades for each strategy.

I also included the TQQQ YTD data above, going back to 2008 (with no momentum overlay just either in TQQQ or TMF) it ran 72.86% CAGR and 47.16% max daily DD. Obviously, if you add momentum overlay here you get a better result. I'm also sure you probably get a worse result if you go unlevered TLT. I don't think I would straight up use TMF as the risk off asset, I would stick to TLT and/or use the dual momentum of TLT vs cash just in case we have a period of stock market weakness that coincides with rising rates.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Fri Nov 27, 2020 3:47 pm

Thanks for posting pm. I need to go back and review his website as it has been quite a while. What he does incorporates a number of things that I use in my own investing/trading and I think they are solid conceptually. One of the great errors it took me way too long in my life to rectify was realizing that in investing/trading "good enough" really is good enough. We humans have an innate desire to tinker in order to improve.

One thing I strongly recommend is that people take a CAGR and just simulate it going forward and see what happens...oftentimes one's financial goals can be easily met with something quite a bit less in terms of performance and at the end of the day that's what it is all about.

One thing on economic data...it is really difficult to deal with; so much so that I don't normally mess with it much unless I absolutely feel a compelling need to. There are two problems with it that can kill you metaphorically speaking when it comes to backtesting. The least problematic is dealing with publishing delays and changes of release dates. You really need to know the ins and outs historically of an economic data stream to use it properly. The second problem is a lot of economic data is revised and there are very few sources (and usually they cost quite a bit) that provide unrevised historical economic data. Having noted the above, I do use it in my current routine. To simplify things I make a very non-elegant, but safe hack. No matter when a series comes into my database it will be recorded as the last day of the month and then I lag it by another month. This introduces its own set of problems but does eliminate forward looking backtests (which will always overstate performance in an optimization).
pmward
Executive Member
Executive Member
Posts: 1731
Joined: Thu Jan 24, 2019 4:39 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by pmward » Fri Nov 27, 2020 4:06 pm

Kbg wrote:
Fri Nov 27, 2020 3:47 pm
Thanks for posting pm. I need to go back and review his website as it has been quite a while. What he does incorporates a number of things that I use in my own investing/trading and I think they are solid conceptually. One of the great errors it took me way too long in my life to rectify was realizing that in investing/trading "good enough" really is good enough. We humans have an innate desire to tinker in order to improve.

One thing I strongly recommend is that people take a CAGR and just simulate it going forward and see what happens...oftentimes one's financial goals can be easily met with something quite a bit less in terms of performance and at the end of the day that's what it is all about.

One thing on economic data...it is really difficult to deal with; so much so that I don't normally mess with it much unless I absolutely feel a compelling need to. There are two problems with it that can kill you metaphorically speaking when it comes to backtesting. The least problematic is dealing with publishing delays and changes of release dates. You really need to know the ins and outs historically of an economic data stream to use it properly. The second problem is a lot of economic data is revised and there are very few sources (and usually they cost quite a bit) that provide unrevised historical economic data. Having noted the above, I do use it in my current routine. To simplify things I make a very non-elegant, but safe hack. No matter when a series comes into my database it will be recorded as the last day of the month and then I lag it by another month. This introduces its own set of problems but does eliminate forward looking backtests (which will always overstate performance in an optimization).
Yeah get back to me once you've dug in. Curious what you will say.

And yeah, I agree on all fronts. Admittedly this is more than I need. It's way more than "good enough". But I'm also blessed to be in a position that I also don't really have anything to lose in trying, so why not? Plus, I enjoy doing this stuff. It brings entertainment in my day to day life. At least this form of entertainment makes me money instead of costing me money, haha.

I also agree with your critique on the backtests of economic data. I think the real nice part of this system is that the vol component is completely separate. The vol component alone backtests really well. So if there is a lag in the economic data, but the market sniffs out economic weakness and starts to sell off, the vol component would hit the eject button for you in advance (and most of the time vol does go red before the SpyComp signal).

Also, as an aside, SpyComp doesn't make any trades based on economic data alone. It's just the trigger that when red defers to the trend following component. He specifically chose some leading economic indicators that would go "red" in advance without the lag effect. For instance, right now 1 of the 7 indicators is still red, so the whole economic portion is red. So currently it is deferred to the trend system, which obviously is green seeing is at we are at all time highs. The 7 indicators he uses are unemployment rate, weekly unemployment claims, retail food and sales, industrial production, yield curve, NFCI leverage indicator, and personal consumption and expenditures. He already made mention in email a couple of times that the next few months of economic data will be kind of skewed green, as the stats from the corona virus set a much lower YoY bar than is typical for some of the indicators.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Sat Nov 28, 2020 12:02 am

I did a very quick backtesting pass on the vol model. You can do something similar by simply using what he is showing on the very first page by going to stockcharts.com. He notes he is using the futures contracts pricing which is not the same as the underlying index...but a similar public version would be his chart. I've done a lot of work in this area previously and I will simply say, it is easy to optimize and think you have something really good when you do not. Simplicity and the basic principles of what backwardization and contango are telling you is the most important thing. All said and done, volatility systems definitely limit big drawdowns very effectively for a modest cost in terms of CAGR over buy and hold. Leveraging up a little when the all clear siren is going off will go a long way to making up for and exceeding lost ground going in and out of the market in response to these signals.
pmward
Executive Member
Executive Member
Posts: 1731
Joined: Thu Jan 24, 2019 4:39 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by pmward » Sat Nov 28, 2020 7:58 am

Kbg wrote:
Sat Nov 28, 2020 12:02 am
I did a very quick backtesting pass on the vol model. You can do something similar by simply using what he is showing on the very first page by going to stockcharts.com. He notes he is using the futures contracts pricing which is not the same as the underlying index...but a similar public version would be his chart. I've done a lot of work in this area previously and I will simply say, it is easy to optimize and think you have something really good when you do not. Simplicity and the basic principles of what backwardization and contango are telling you is the most important thing. All said and done, volatility systems definitely limit big drawdowns very effectively for a modest cost in terms of CAGR over buy and hold. Leveraging up a little when the all clear siren is going off will go a long way to making up for and exceeding lost ground going in and out of the market in response to these signals.
Yep exactly, I used stockcharts.com as well to backtest in my own research. I do think the vol component is the most important piece to this puzzle. For many reasons vol is the tail that wags the dog these days. Most flows are systematic, and a good chunk of systematic flows are based on vol. When vol is going down, systematic flows go into the market, when vol is going up systematic flows come out of the market. This system is basically designed to follow along with those flows.

I view SpyComp not as the main course, but as the backup plan in case some strange unforeseen occurrence happens to fool the vol markets, then you have the fallback of some economic leading indicators and a simple trend system to get you out of the way.

Then you have the relative momentum overlay and leverage to juice up returns when you have the all clear sign. Not every individual trade is a winner, but it will capture the big secular moves and trends, which is where most of the money is made anyways, and avoid the big secular bear markets and deep corrections, which is where most money is lost. I personally have confidence that this will continue to outperform SPY by a large enough clip going forward to justify using this model even in a taxable account.

I think it is a very intelligently designed system as a whole. And like I mentioned in the OP, it's one that is easily customizable, which I have done myself. One doesn't have to perfectly reverse engineer Paul's proprietary model to still get similar returns. Just using the info he has given one can easily get "close enough". Even if one just took the open source GTT trend model instead of SpyComp linked in the OP, and just used simple backwardation or contango of the vol curve as the signal without the moving averages Paul uses, along with a simple momentum overlay I think they would still do really well.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Sat Nov 28, 2020 9:23 am

Forgot to add, the CBOE site allows you to download historical data for most of the various VIX indices if one wants to backtest.

http://www.cboe.com/index/

The symbols in time order are...VIX9D, VIX, VIX3M, VIX6M, VIX1Y.

And for the PP enthusiast gold, and TLT have their own volatility indexes.

Good on CBOE, they are one of the view index providers that doesn't charge for historical data anymore.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Sat Nov 28, 2020 10:17 am

One more source...the site is not actively maintained/posted to anymore but if one is interested in GAA type strategies this has several of them.

https://indexswingtrader.blogspot.com/p ... gnals.html

The "picks" section appears to be still working via automation quietly in the background.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Sat Nov 28, 2020 10:50 am

Readings on dual momentum...read both the original and the rebuttal

Original: https://blog.thinknewfound.com/2019/01/ ... entum-gem/

Rebuttal: https://dualmomentum.net/2019/01/17/whi ... entum-gem/

You will learn a TON by reading these two articles. I was persuaded by the original and made adjustments to my approach accordingly. In practice what I have found using segmented momentum ranging from 3-12 months is that transitions between assets are no longer binary. It's more gradual in and out and the way I do it I can be holding 1-3 of the assets at the same time. Sticking with classic dual momentum is much easier to execute. What I do requires programming and data to pull off.
pmward
Executive Member
Executive Member
Posts: 1731
Joined: Thu Jan 24, 2019 4:39 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by pmward » Sat Nov 28, 2020 11:27 am

Kbg wrote:
Sat Nov 28, 2020 10:50 am
Readings on dual momentum...read both the original and the rebuttal

Original: https://blog.thinknewfound.com/2019/01/ ... entum-gem/

Rebuttal: https://dualmomentum.net/2019/01/17/whi ... entum-gem/

You will learn a TON by reading these two articles. I was persuaded by the original and made adjustments to my approach accordingly. In practice what I have found using segmented momentum ranging from 3-12 months is that transitions between assets are no longer binary. It's more gradual in and out and the way I do it I can be holding 1-3 of the assets at the same time. Sticking with classic dual momentum is much easier to execute. What I do requires programming and data to pull off.
Thanks, I'm about to head out the door but I will read these later and report back. Also, for any momentum strategy I don't do the traditional 12 month only look back. I like to use the average of 3, 6, and 12 month returns. Backtesting works much better than a single time period. It helps you get into new trends and out of old trends earlier.

By the way, thanks for actual non-political investment discussion. My sanity needs this right now, haha.
Kbg
Executive Member
Executive Member
Posts: 1847
Joined: Fri May 23, 2014 4:18 pm

Re: Combining Vol, Economic, Trend, and Momentum Strategies

Post by Kbg » Sat Nov 28, 2020 12:39 pm

On economic signals...I use a combination of various macro data series and an old fashioned monthly moving average.

The code for the signal is: IIf(MacroEconSig >= 0, 1, IIf(MacroEconSig < 0 AND C > MA(C,6), 1, -1));

In English:

If the MacroEcon signal is >=0 then = 1
If the MacroEcon signal is < 0 then check to see if the monthly close is above the 6 month moving average if it is then also =1. If not, then -1.

1 is long, -1 is flat. Since 1/1/2008 the CAGR is 15 and the MDD is -20

I use 7 different macro econ series that are either +1 or -1 at all times.
Post Reply