HEDGEFUNDIE's excellent adventure Part II: The next journey

[mroe800]

Today’s a good rebalance day. Much better than when I rebalanced last week.

[rchmx1]

Very glad I bought puts on TMF! As I said earlier, I see very few ways this ETF won't be down over the next few years, even if rates fall. I suspect it's going to be down 30% or more. I also think the correlation between treasuries and stocks will be weaker than ever, due to bonds being less a safe haven at these rates. But as long as we don't see a stock market crash, those running the Hedgefundie strategy probably won't see drawdowns larger than 40% or so.

Smart move! I'm glad I sold around 90% of the TMF I was holding when it was around the $42 mark, it seemed like this was the way it was headed. I really need to finish the book on options I've started reading, so I can take advantage of this type of insurance.

[langlands]

My biggest concern right now is TMF's yields combined with absurdly high financing costs. I am waiting for the annual report of TMF, coming around October 31st. I want to see just exactly how much interest they are paying on swaps.

The April report said that they were paying apparently 1.2% in interest to borrow for TMF; which is absolutely ludicrous when the LIBOR rate was 0.37% that day. Keep in mind that TMF borrows >250% of its value, only about 50% are in direct long term treasuries in $TLT (which are the collateral).

If you are in UPRO/TQQQ; you don't have to worry. Their borrowing costs were far more reasonable, at like 0.8% or so on the same day.

Furthermore, TMF's swaps are a leverage on the $TLT ETF fund; with a ER of 0.15%. I do not believe this is reported in the TMF SEC ER because they can technically say that it's not an acquired fund expense, they're just buying swaps on the TLT index and it just turns out the TLT index is deflated by TLT ERs.

Looking at the fund fact sheet, they say acquired fund expenses are 0.13%, which is clearly too low for 0.15% x 3 (due to the 3x leverage). So, it seems pretty clear that they are only counting the direct TLT holdings' acquired expenses; and the TLT ER deflator is an additional (undisclosed) drag on returns.

Anyways, let's look at all the fees you are paying with TMF:

• 0.15% * 2.5 in the form of TLT ER tracking drag on swaps = 0.375%
  1.2% * 2.5 in the form of swap financing premiums = 3%
  1.05% in the form of TMF ER = 1.05%

Total real expense ratio of TMF = 4.425% + volatility drag

I don't care about historical performance, when 30 year treasuries were paying 3.45% (Jan 2019 ), 10.35% - (4.425% + libor adjustment) - vol drag = positive.

But now with a yield to maturity of 1.34%, 4.02% - 4.425% - vol drag, it would be an absolute no go. It is an asset with a negative expected value, in nominal terms even, and crazy stdev. You never invest in things with negative expected values: that is called losing money.

I am just hoping the 1.2% swap financing costs (which you pay ~2.5x) was a fluke on that particular day (April 30th) the report was generated, and real financing costs have came down dramatically. Otherwise, I will be replacing TMF.

Hi coingaroo, could you link the documents where you find the financing costs for TMF, UPRO, and TQQQ? You seem to think TMF (1.2%) is more expensive to borrow than UPRO and TQQQ (0.8%) whereas Uncorrelated and others seem to think it's the reverse.

Thanks.

[Uncorrelated]

Hi coingaroo, could you link the documents where you find the financing costs for TMF, UPRO, and TQQQ? You seem to think TMF (1.2%) is more expensive to borrow than UPRO and TQQQ (0.8%) whereas Uncorrelated and others seem to think it's the reverse.

Thanks.

For what it's worth, I agree that coingaroo's analysis is accurate at the moment the last annual report was released. However, I suspect that the increase in swap fees was a temporarily blip, in particular because proshares treasury ETF's show more reasonable numbers.

But until we get access to the next round of annual reports, there is no way to know what the current swap rates are.

[langlands]

Hi coingaroo, could you link the documents where you find the financing costs for TMF, UPRO, and TQQQ? You seem to think TMF (1.2%) is more expensive to borrow than UPRO and TQQQ (0.8%) whereas Uncorrelated and others seem to think it's the reverse.

Thanks.

For what it's worth, I agree that coingaroo's analysis is accurate at the moment the last annual report was released. However, I suspect that the increase in swap fees was a temporarily blip, in particular because proshares treasury ETF's show more reasonable numbers.

But until we get access to the next round of annual reports, there is no way to know what the current swap rates are.

I found the TMF annual report from October 31 2019, which is 304 pages long. Is there an easy way to find where this information is? I tried Crtl + F some obvious search terms and couldn't locate it.

[Uncorrelated]

I found the TMF annual report from October 31 2019, which is 304 pages long. Is there an easy way to find where this information is? I tried Crtl + F some obvious search terms and couldn't locate it.

The annual reports for TMF can be found here: https://www.sec.gov/cgi-bin/browse-edga ... C000075253
The relevant documents are N-CSRS (certified shareholder report semi-annual), N-CSR (certified shareholder report) and N-Q (quarterly holdings report).

The obvious search terms are the name of the ETF, "LIBOR", or "Long Total Return Swap Contracts". In the latest N-CSRS (click the first .htm document), this information is detailed on page 91:



The relevant LIBOR rate can be inspected on FRED, which was 0.32% at april 30.


We can also observe the rates from a similar proshares ETF also on edgar.



The 1-month LIBOR rate on May 30 was 0.18%.

[Steve Reading]

I found the TMF annual report from October 31 2019, which is 304 pages long. Is there an easy way to find where this information is? I tried Crtl + F some obvious search terms and couldn't locate it.

The annual reports for TMF can be found here: https://www.sec.gov/cgi-bin/browse-edga ... C000075253
The relevant documents are N-CSRS (certified shareholder report semi-annual), N-CSR (certified shareholder report) and N-Q (quarterly holdings report).

The obvious search terms are the name of the ETF, "LIBOR", or "Long Total Return Swap Contracts". In the latest N-CSRS (click the first .htm document), this information is detailed on page 91:



The relevant LIBOR rate can be inspected on FRED, which was 0.32% at april 30.


We can also observe the rates from a similar proshares ETF also on edgar.



The 1-month LIBOR rate on May 30 was 0.18%.

So the spread over LIBOR at which TMF borrowed was ~1%? Am I getting that right?

[langlands]

I found the TMF annual report from October 31 2019, which is 304 pages long. Is there an easy way to find where this information is? I tried Crtl + F some obvious search terms and couldn't locate it.

The annual reports for TMF can be found here: https://www.sec.gov/cgi-bin/browse-edga ... C000075253
The relevant documents are N-CSRS (certified shareholder report semi-annual), N-CSR (certified shareholder report) and N-Q (quarterly holdings report).

Thanks a lot.

[Uncorrelated]

So the spread over LIBOR at which TMF borrowed was ~1%? Am I getting that right?

Yes. This represents a total expense ratio for TMF in the range of 3.5%.

I believe this was a temporarily blip caused by the COVID-19 crisis, as evidenced by the fact that proshares' swap rates 1 month later were substantially lower (near zero). But we have no way to know for sure.

[occambogle]

Anyone know why yields on LTTs are going up so much (and prices of TMF,EDV, etc down so much) this week? Is it just "risk-on" or something more closely linked to rates/inflation expectations? TMF dropped more than 6% yesterday, and from my casual observing that seems a lot more than just a move in response to equities rising.... but I've no idea.

[kim.gold]

Anyone know why yields on LTTs are going up so much (and prices of TMF,EDV, etc down so much) this week? Is it just "risk-on" or something more closely linked to rates/inflation expectations? TMF dropped more than 6% yesterday, and from my casual observing that seems a lot more than just a move in response to equities rising.... but I've no idea.

Your assessment is probably correct https://www.cnbc.com/2020/09/30/treasur ... focus.html

“The traditional correlation between higher equities and yields appears to be functioning again; at least in the very near-term,” Ian Lyngen, head of U.S. rates at BMO, said in a note on Tuesday.

[kjm]

Anyone know why yields on LTTs are going up so much (and prices of TMF,EDV, etc down so much) this week? Is it just "risk-on" or something more closely linked to rates/inflation expectations? TMF dropped more than 6% yesterday, and from my casual observing that seems a lot more than just a move in response to equities rising.... but I've no idea.

I assume this is happening in anticipation of a stimulus package, which will have to be paid for by issuing new debt. It's going to be interesting to see whether the market can absorb the new debt without the Fed having to step up its bond buying program.

[chrisdds98]

Anyone know why yields on LTTs are going up so much (and prices of TMF,EDV, etc down so much) this week? Is it just "risk-on" or something more closely linked to rates/inflation expectations? TMF dropped more than 6% yesterday, and from my casual observing that seems a lot more than just a move in response to equities rising.... but I've no idea.

I assume this is happening in anticipation of a stimulus package, which will have to be paid for by issuing new debt. It's going to be interesting to see whether the market can absorb the new debt without the Fed having to step up its bond buying program.

are they making progress on stimulus talks? Last i heard it was just going to be an airline bailout. heard they are still far apart on other stimulus

[gokuisthebest]

[@perfectuncertainty, could you perhaps have a quick look and verify the calculated results (for min var and inv vol) look alright to you? I more or less copied the logic from your file but just want to be sure as this is not my area of expertise.

The calculations are correct for the samples I tried.

I have a few questions about AAA min variance.
1. Can I know when to calculate and when to rebalance? start/end of month & start/end of day.
2. Should i start only on the 1st of a month or is any day fine and stick with the plan?
3. What about a swing on the day(or the day before) of rebalance? wouldnt it affect the entire month allocation?(i understand this can happen even for HFEA)

im thinking these small changes shouldnt affect in the long run as they would even out but i would be glad if you can tell me these details for the backtests.
Thank you.

[BuffMaltese]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

[gokuisthebest]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

I see, thanks. 8% seems too low which could happen multiple times a month.
Can I know when you started with AAA MV? I agree currently it shows high value for TMF(~70%). Could be higher or much lower next month.
Since PV allows only monthly values, in this AAA MV, there were few times where TMF had some wide swings from month to next. I think to see the benefit of AAA, one needs atleast ~3-5 yrs.

[BuffMaltese]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

I see, thanks. 8% seems too low which could happen multiple times a month.
Can I know when you started with AAA MV? I agree currently it shows high value for TMF(~70%). Could be higher or much lower next month.
Since PV allows only monthly values, in this AAA MV, there were few times where TMF had some wide swings from month to next. I think to see the benefit of AAA, one needs atleast ~3-5 yrs.

PV gives a daily signal when you are under the trial period and I’m assuming if you pay. There is an excel link in this tread too.

Just started near the end of sept (terrible timing I guess) I believe it was signaling around 80% tmf which was concerning to me because looking back the last 10 years it’s the highest recommended allocation ever and there has been seemingly unrelenting news regarding selling tlt and/or tmf for a myriad of reasons. Basically took an immediate large loss which was particularly unsettling to me because it was due to bonds/treasuries and the market hadn’t really done anything that unusual otherwise.

[rchmx1]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

I see, thanks. 8% seems too low which could happen multiple times a month.
Can I know when you started with AAA MV? I agree currently it shows high value for TMF(~70%). Could be higher or much lower next month.
Since PV allows only monthly values, in this AAA MV, there were few times where TMF had some wide swings from month to next. I think to see the benefit of AAA, one needs atleast ~3-5 yrs.

PV gives a daily signal when you are under the trial period and I’m assuming if you pay. There is an excel link in this tread too.

Just started near the end of sept (terrible timing I guess) I believe it was signaling around 80% tmf which was concerning to me because looking back the last 10 years it’s the highest recommended allocation ever and there has been seemingly unrelenting news regarding selling tlt and/or tmf for a myriad of reasons. Basically took an immediate large loss which was particularly unsettling to me because it was due to bonds/treasuries and the market hadn’t really done anything that unusual otherwise.

That's a hell of a test early on for your stomach for this as a long term strategy. I also plugged in my ~40% TMF allocation for the first time in August when TMF got up to $45-46. Near immediate regret when it quickly dropped to ~$40, sold most if it a short time later when it rose back to around $42, as an opportunity to buy one of the major recent dips in UPRO/TQQ. Already made back more than the money I lost buying TMF at its peak. One of the things that keeps ringing in my ears after reading through these two monster threads is that the market situation when you start this strategy has an outsized impact on its outcome. That seems to be relevant for TMF as much as UPRO.

[BuffMaltese]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

I see, thanks. 8% seems too low which could happen multiple times a month.
Can I know when you started with AAA MV? I agree currently it shows high value for TMF(~70%). Could be higher or much lower next month.
Since PV allows only monthly values, in this AAA MV, there were few times where TMF had some wide swings from month to next. I think to see the benefit of AAA, one needs atleast ~3-5 yrs.

PV gives a daily signal when you are under the trial period and I’m assuming if you pay. There is an excel link in this tread too.

Just started near the end of sept (terrible timing I guess) I believe it was signaling around 80% tmf which was concerning to me because looking back the last 10 years it’s the highest recommended allocation ever and there has been seemingly unrelenting news regarding selling tlt and/or tmf for a myriad of reasons. Basically took an immediate large loss which was particularly unsettling to me because it was due to bonds/treasuries and the market hadn’t really done anything that unusual otherwise.

That's a hell of a test early on for your stomach for this as a long term strategy. I also plugged in my ~40% TMF allocation for the first time in August when TMF got up to $45-46. Near immediate regret when it quickly dropped to ~$40, sold most if it a short time later when it rose back to around $42, as an opportunity to buy one of the major recent dips in UPRO/TQQ. Already made back more than the money I lost buying TMF at its peak. One of the things that keeps ringing in my ears after reading through these two monster threads is that the market situation when you start this strategy has an outsized impact on its outcome. That seems to be relevant for TMF as much as UPRO.

Yeah, I was surprised how much it was affecting me. I was waking up early and becoming overly preoccupied watching the market during the day having difficulty working.

I really didn’t want to sell any of the loss in any of my taxable accounts till the end of the month but I just couldn’t stand it anymore and sold a large chunk of tmf and immediately bought sso with the money. I was literally stunned in disbelief having apparently made the move seconds before the market reacted to Trump’s tweet sending tmf +3% and spy free falling compounding my losses. Really just unbelievable.

[babbysfirststonk]

Have you tried adding VIX to the portfolio?

Here's a really quick backtest I ran from as far back as I could go until Sep 31, 2020. Adding 5% VIX increases your sharpe and sortino ratios, decreases your max drawdown throughout the market shock, and allows you to increase your UPRO percentage because of the added risk parity.

Let me know what you guys think. I currently have 5% VIXM and 5% XVZ in my personal portfolio which runs a similar strategy to @Hedgefundie.

[privatefarmer]

This strategy, like any other, has to be for the long term for it to work. Last month my min variance strategy had me 100% in TQQQ/DRN (REITS), I ended up down about 15% but was down even more intra-month before it recovered some. This month, it has me in 80% TMF/20% DRN.

The reality is, if you backtest a variant of this strategy going back 30+ Years (looking at the unleveraged counterparts), it’s had a remarkably CONSISTENT return/steady climb up. It has had down years, of course, but has been far more consistent than holding solely stocks. If using LETFs, the unleveraged counterparts only need to return a small % in order for the LETFs to have a strong return. I think looking From 5000 feet, a risk parity / minimum variance strategy involving as many different risk assets as you can is a great long term strategy but it has to be long term and you have to be able to stomach stretches of under performance.

[perfectuncertainty]

[@perfectuncertainty, could you perhaps have a quick look and verify the calculated results (for min var and inv vol) look alright to you? I more or less copied the logic from your file but just want to be sure as this is not my area of expertise.

The calculations are correct for the samples I tried.

I have a few questions about AAA min variance.
1. Can I know when to calculate and when to rebalance? start/end of month & start/end of day.
2. Should i start only on the 1st of a month or is any day fine and stick with the plan?
3. What about a swing on the day(or the day before) of rebalance? wouldnt it affect the entire month allocation?(i understand this can happen even for HFEA)

im thinking these small changes shouldnt affect in the long run as they would even out but i would be glad if you can tell me these details for the backtests.
Thank you.

That's a personal choice. I believe most rebalance monthly (time-based). Others would use banded rebalancing based on percentage change from prior allocation.

Uncorrelated (another user on here) might have info to share on specific days to rebalance if you are time-based. I'd recommend keeping it simple, whichever you choose.

[coingaroo]

Agree with all the thoughts here that this is a long term strategy. I am actually not concerned about the drop in TMF values: yields are going up and as long as it's not a lot of see-sawing (vol decay). Ultimately this means more crash protection and more returns for a long term investor, as higher yields = higher bond returns.

TLT (the ETF that TMF is based on) is at 1.5% YTM. That's still so low compared to historical levels, but it's modestly better than 1.2%-ish we were seeing a while ago. Should yields continue to go up, I welcome it.

[RovenSkyfall]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

This brings up something I have been thinking about a lot lately. Does anyone know the start date sensitivity of this strategy? I.e. should people DCA into this strategy because the start date sensitivity is so large? From my understanding, with regular index investing, people are better off 2/3 of the time with lump sum investments compared to DCA, but is this strategy different?

In a lot of the back tests, CAGR is compared from a single start date. If it is very start date specific, is there a more reliable way to look at the potential outcomes given the start date sensitivity (i.e. MC or rolling returns)? Should AAA MV, quarterly re-balancing and IV be compared with regards to rolling returns (or MC) rather than start and end date CAGRs to give a more reliable description of possible returns (rather than just looking at the longest data set)?

Thank you in advance to those who understand the metrics better.

[gokuisthebest]

I use TQQQ, QLD, QQQ, TMF and TLT and put it into Inverse Volatility. Deleveraging with QQQ and TLT suits me.

Can you tell me what you mean by this and how you achieve it? how is qqq deleveraging this portfolio when you have tqqq&qld.

[ljford7]

This brings up something I have been thinking about a lot lately. Does anyone know the start date sensitivity of this strategy? I.e. should people DCA into this strategy because the start date sensitivity is so large? From my understanding, with regular index investing, people are better off 2/3 of the time with lump sum investments compared to DCA, but is this strategy different?

In a lot of the back tests, CAGR is compared from a single start date. If it is very start date specific, is there a more reliable way to look at the potential outcomes given the start date sensitivity (i.e. MC or rolling returns)? Should AAA MV, quarterly re-balancing and IV be compared with regards to rolling returns (or MC) rather than start and end date CAGRs to give a more reliable description of possible returns (rather than just looking at the longest data set)?

Thank you in advance to those who understand the metrics better.

Rolling returns is a good metric to look at. Also, do some of your own backtests in PV and try to break the strategy (pick days to start when the market crashed or randomly pick days (you will be surprised at the results).

I looked really hard at it (UPRO/TMF) and at the end of the day, even if it doesn't make a ton of money, but keeps decent pace with the S&P, I will be happy that I tried it. If it does make a bunch of money, then it might enable me to retire earlier (worth the gamble in my opinion).

[langlands]

Idk, I couldn’t get myself to stick with MV, couldn’t of been further off. Absolutely loaded on TMF just in time for the drop. Really wish I started with the standard AA. To answer your question, I think something like an 8% fluctuation from the monthly signal warrants an adjustment.

I wouldn’t think the day started would matter, just avoiding a wash sale in taxable.

It’s really hard to leave alone for a month at a time when the signal changes drastically and it also happens to align with your gut. I don’t think it’s for me

This brings up something I have been thinking about a lot lately. Does anyone know the start date sensitivity of this strategy? I.e. should people DCA into this strategy because the start date sensitivity is so large? From my understanding, with regular index investing, people are better off 2/3 of the time with lump sum investments compared to DCA, but is this strategy different?

In a lot of the back tests, CAGR is compared from a single start date. If it is very start date specific, is there a more reliable way to look at the potential outcomes given the start date sensitivity (i.e. MC or rolling returns)? Should AAA MV, quarterly re-balancing and IV be compared with regards to rolling returns (or MC) rather than start and end date CAGRs to give a more reliable description of possible returns (rather than just looking at the longest data set)?

Thank you in advance to those who understand the metrics better.

I don't think DCA vs. lump sum considerations are any different for this strategy. It is simply more volatile, but that doesn't affect the decision.

Don't take this the wrong way, but I think that all the detailed backtests being posted recently are completely useless and misleading. The truth is that if you follow this strategy faithfully for 30 years, your ending balance can be anywhere from 2x to 100x your starting amount.

[perfectuncertainty]

I use TQQQ, QLD, QQQ, TMF and TLT and put it into Inverse Volatility. Deleveraging with QQQ and TLT suits me.

Can you tell me what you mean by this and how you achieve it? how is qqq deleveraging this portfolio when you have tqqq&qld.

Let's say you have $100k and you choose TQQQ and TMF. You can follow the 55/45 HFEA allocation or you can use AAA for a risk parity approach based on Inverse Volatility or Minimum Variance, or you can use a Target Volatility approach.

For AAA you don't have a way to deleverage except to leave money in cash constantly. I personally lever down by choosing the AAA Inverse volatility method so that QQQ and TLT end up with larger percentages versus putting it in cash. I sleep better that way - a personal choice.

Here is the PV I use for this: Link

[occambogle]

Let's say you have $100k and you choose TQQQ and TMF. You can follow the 55/45 HFEA allocation or...

Maximum Sharpe for TMF/TQQQ is a little less than 55/45, more like 52/48. I usually aim for that when rebalancing but I imagine the difference is marginal and TQQQ usually ends up above that pretty quickly.

The continued volatility this year seems to really affect results though, presumably through volatility decay of the daily-leverage ETFs....

May to Oct 2020 PV link

[babbysfirststonk]

Actually, someone already wrote about the VIX strategy I mentioned above: https://earlyretirementnow.com/2020/04/ ... k-in-2020/

I'd still like to hear feedback though. It's strategy #2 in that link.

[kjm]

Actually, someone already wrote about the VIX strategy I mentioned above: https://earlyretirementnow.com/2020/04/ ... k-in-2020/

I'd still like to hear feedback though. It's strategy #2 in that link.

I don't think a permanent position in VXX is the best way to improve risk-adjusted returns. It takes more than it gives over the long run. Now, if you could figure out how to 1) hold it strategically and 2) size the position appropriately, you'd be onto something.

If the goal is to improve risk-adjusted returns, the following strategy is effective and low-effort.

TQQQ weight = average standard deviation / trailing 15-day standard deviation
TMF weight = average standard deviation / trailing 15-day standard deviation

Then, once you have your allocation, you come up with a leverage factor by running the same equation over the entire portfolio. Here are the results I get going back 10 years.

MODEL RESULTS
Annualized Return: 40.79%
Max Drawdown: -23.14%
Standard Deviation: 23.75%

55/45 RESULTS
Annualized Return: 43.33%
Max Drawdown: -40.09%
Standard Deviation: 29.73%

TQQQ RESULTS
Annualized Return: 44.55%
Max Drawdown: -69.92%
Standard Deviation: 57.86%

TMF RESULTS
Annualized Return: 15.78%
Max Drawdown: -53.51%
Standard Deviation: 43.6%

[Uncorrelated]

Actually, someone already wrote about the VIX strategy I mentioned above: https://earlyretirementnow.com/2020/04/ ... k-in-2020/

I'd still like to hear feedback though. It's strategy #2 in that link.

MODEL RESULTS

10 years is not nearly enough data to draw conclusions, backtests are not suited for this type of analysis.

I summarized some academic literature on volatility timing in this post. Basically, the consensus seems to be that previous papers in support of volatility timing contain methodological errors and data snooping biases. After correcting for those errors, these strategies don't work.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

[RovenSkyfall]

Actually, someone already wrote about the VIX strategy I mentioned above: https://earlyretirementnow.com/2020/04/ ... k-in-2020/

I'd still like to hear feedback though. It's strategy #2 in that link.

MODEL RESULTS

10 years is not nearly enough data to draw conclusions, backtests are not suited for this type of analysis.

I summarized some academic literature on volatility timing in this post. Basically, the consensus seems to be that previous papers in support of volatility timing contain methodological errors and data snooping biases. After correcting for those errors, these strategies don't work.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

Do these statements include AAA with minimum variance?

[coingaroo]

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

Do simplistic timing methods you've tested include trend following (e.g. 200 day SMA)? That seems to be the (only?) commonly known timing method with thorough evidence behind it. Of course, it didn't work for the COVID crash, but trend following losing to whipsaws is well known.

[kjm]

10 years is not nearly enough data to draw conclusions, backtests are not suited for this type of analysis.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

The challenge is we only have about 10 years of data for these leveraged ETFs. I ran the same test on QQQ and TLT going back to 2002 and got the following. I'd go further, but I can't find historical data for longer duration bonds.

MODEL RESULTS
Annualized Return: 9.98%
Max Drawdown: -14.89%
Standard Deviation: 8.68%

55/45 RESULTS
Annualized Return: 11.08%
Max Drawdown: -27.81%
Standard Deviation: 11.29%

QQQ RESULTS
Annualized Return: 15.64%
Max Drawdown: -53.55%
Standard Deviation: 21.98%

TLT RESULTS
Annualized Return: 3.22%
Max Drawdown: -28.46%
Standard Deviation: 13.96%

Again, better risk-adjusted returns for the target volatility approach.

I summarized some academic literature on volatility timing in this post. Basically, the consensus seems to be that previous papers in support of volatility timing contain methodological errors and data snooping biases. After correcting for those errors, these strategies don't work.

The abstract mentions volatility-managed portfolios don't "outperform" unmanaged portfolios. Absolute out-performance isn't what I'm after here. I'm trying to reduce the risk profile (standard deviation and max drawdown).

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

I agree you can get better absolute and risk-adjust returns by incorporating multiple factors. Just trying to illustrate a point.

Just want to add things get better when you add leverage. If I cap leverage at 2x rather than 1x (as above), I get the following.

MODEL RESULTS
Annualized Return: 13.26%
Max Drawdown: -15.14%
Standard Deviation: 10.96%

Using both vol targeting + leverage is really the sweet spot.

[willthrill81]

10 years is not nearly enough data to draw conclusions, backtests are not suited for this type of analysis.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

The challenge is we only have about 10 years of data for these leveraged ETFs.

Others have constructed datasets for how these ETFs would have been likely to perform going back decades. I'm not sure how much of that is included in this thread, but I know that there was a lot of that discussion in the first thread, which was referenced in the OP of this thread.

[Uncorrelated]

Do these statements include AAA with minimum variance?

I have not tested that specific strategy because the theoretical basis behind it is laughable.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

Do simplistic timing methods you've tested include trend following (e.g. 200 day SMA)? That seems to be the (only?) commonly known timing method with thorough evidence behind it. Of course, it didn't work for the COVID crash, but trend following losing to whipsaws is well known.

You will be interested in this paper: https://papers.ssrn.com/sol3/papers.cfm ... id=3165284.

The authors argue that there is no evidence (in any asset class) that time series momentum predict future returns. They show that statistical models that previous papers have used to investigate momentum are flawed. They also attempt to isolate the return-predictability element of time series momentum and the volatility timing aspects of momentum strategies, and show that favorable backtests are not due to the strategies ability to predict returns, but rather due to accidental volatility timing aspects. If you have been using trend following strategies, all the improved risk-adjusted returns can be explained with volatility managed strategies.

I suspected that this was the case for a long time, but it was very difficult to find papers investigating it.

10 years is not nearly enough data to draw conclusions, backtests are not suited for this type of analysis.

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

The challenge is we only have about 10 years of data for these leveraged ETFs. I ran the same test on QQQ and TLT going back to 2002 and got the following. I'd go further, but I can't find historical data for longer duration bonds.

MODEL RESULTS
Annualized Return: 9.98%
Max Drawdown: -14.89%
Standard Deviation: 8.68%

55/45 RESULTS
Annualized Return: 11.08%
Max Drawdown: -27.81%
Standard Deviation: 11.29%

QQQ RESULTS
Annualized Return: 15.64%
Max Drawdown: -53.55%
Standard Deviation: 21.98%

TLT RESULTS
Annualized Return: 3.22%
Max Drawdown: -28.46%
Standard Deviation: 13.96%

Again, better risk-adjusted returns for the target volatility approach.

I summarized some academic literature on volatility timing in this post. Basically, the consensus seems to be that previous papers in support of volatility timing contain methodological errors and data snooping biases. After correcting for those errors, these strategies don't work.

The abstract mentions volatility-managed portfolios don't "outperform" unmanaged portfolios. Absolute out-performance isn't what I'm after here. I'm trying to reduce the risk profile (standard deviation and max drawdown).

This is consistent with my own experiments where I found that simplistic timings methods don't improve expected utility out-of-sample. I was only able to obtain positive out-of-sample performance by combining multiple complicated models. Even then I'm not sure if the results are economically significant.

I agree you can get better absolute and risk-adjust returns by incorporating multiple factors. Just trying to illustrate a point.

Just want to add things get better when you add leverage. If I cap leverage at 2x rather than 1x (as above), I get the following.

MODEL RESULTS
Annualized Return: 13.26%
Max Drawdown: -15.14%
Standard Deviation: 10.96%

Using both vol targeting + leverage is really the sweet spot.

If you're limited to 10 years of data, then honestly the best idea is to give up. There is no chance you will find anything statistically significant with 10 years of data.

The paper says volatility managed strategies have lower certainty equivalent returns. That is academic speak for "lower risk adjusted returns". This is the same method that I have used in my experiments.

It is trivial to beat a constant allocation in-sample, as your backtests show. But this is not what you want, you want to beat a constant allocation out-of-sample. I appreciate the effort but these backtests are completely useless in the best case, and highly misleading in the worst case.

[babbysfirststonk]

I don't think a permanent position in VXX is the best way to improve risk-adjusted returns. It takes more than it gives over the long run. Now, if you could figure out how to 1) hold it strategically and 2) size the position appropriately, you'd be onto something.

If the goal is to improve risk-adjusted returns, the following strategy is effective and low-effort.

I agree about VXX, but keep in mind XVZ has a lot less drawdown. The lowest it ever fell from its initial value when it launched in Sept 2011 was 32%. I bring it up because adding even 5% to the portfolio decreases maximum drawdown at the cost of lowered returns. Here's a backtest showing what I'm talking about. VIX positively correlates with having better worst years, smaller maximum drawdowns, and worse total returns. It's a tradeoff that the investor can decide if they want to make. IMO though, the optimal VIX allocation will probably be small, but nonzero.

And if you use the portfolio optimizer to minimize maximum drawdown over the same time interval, it will actually put 1/3 of your assets in XVZ.

Also, that strategy looks interesting, but just to be clear, that's an active portfolio right?

[RovenSkyfall]

Do these statements include AAA with minimum variance?

I have not tested that specific strategy because the theoretical basis behind it is laughable.

I have always appreciated your posts. Can you educate me on your opinions here?

If you're limited to 10 years of data, then honestly the best idea is to give up. There is no chance you will find anything statistically significant with 10 years of data.

The paper says volatility managed strategies have lower certainty equivalent returns. That is academic speak for "lower risk adjusted returns". This is the same method that I have used in my experiments.

It is trivial to beat a constant allocation in-sample, as your backtests show. But this is not what you want, you want to beat a constant allocation out-of-sample. I appreciate the effort but these backtests are completely useless in the best case, and highly misleading in the worst case.

By this do you mean that volatility/AAA's only appear work in backtests in the exact historical subset they were tested in, but have not been proven to increase returns over constant allocations when adjusted for risk?

[Uncorrelated]

Do these statements include AAA with minimum variance?

I have not tested that specific strategy because the theoretical basis behind it is laughable.

I have always appreciated your posts. Can you educate me on your opinions here?

I think it's very important to have a proper scientific process. First, you should come up with something (an utility function) you want to maximize, for example risk-adjusted return according to some measure. Then you can create a hypothesis of something that is exploitable, for example "market trend predicts future returns". Then you can run statistical tests to see if that hypothesis is true, and then you can design and test the strategy.

Minimum variance fails on all four counts: there is a link between common utility functions and minimum variance, but that utility function categorically disqualifies leveraged products. There is no hypothesis, what predictable market behavior is this strategy supposed to exploit? Obviously no statistical tests have been performed. I have not seen any valid tests of this strategy.

If you're just trying random stuff in a backtest, you are guaranteed to find something that will work. But will it work out-of-sample? You don't know, because your process made it impossible to evaluate that.

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

If you're limited to 10 years of data, then honestly the best idea is to give up. There is no chance you will find anything statistically significant with 10 years of data.

The paper says volatility managed strategies have lower certainty equivalent returns. That is academic speak for "lower risk adjusted returns". This is the same method that I have used in my experiments.

It is trivial to beat a constant allocation in-sample, as your backtests show. But this is not what you want, you want to beat a constant allocation out-of-sample. I appreciate the effort but these backtests are completely useless in the best case, and highly misleading in the worst case.

By this do you mean that volatility/AAA's only appear work in backtests in the exact historical subset they were tested in, but have not been proven to increase returns over constant allocations when adjusted for risk?

The backtests show that the strategy has outperformed in the past, but that doesn't tell you anything about future performance. Using these backtests to make investment decisions is just as valid as putting all your money in last years biggest winner (amazon?).

To illustrate, I run my market timing experiments with k-fold cross validation. I partition the data into K parts (say, 10 parts). One part is selected as the holdout part, the rest of the parts are used to select the optimal parameters. The strategy is then tested on the holdout part. With this testing methodology, I looked at various simple volatility managed strategies and none of them outperform a constant allocation by anything meaningful. This is a simplified version of what they show in the papers I linked a few posts back.

[Tingting1013]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

[hilink73]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

And that is why?

[Jdahlen]

$100,000 will be available for trading on Weds.

I’m looking to enter some long positions on some ETF’s and Mutual Funds.

Assuming I pick some winning strategies with positive upsides and the occasional rebalancing, what’s my potential growth looking like at around the 10 year mark?

It’s an inherited IRA, so I’d like to grow it as much as possible and not take any RMD’s until it’s well above the 300-400K mark.

[Uncorrelated]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

One of the assumptions is allocating 100% of your portfolio to this strategy.

If you bucketize HFEA away from your main portfolio, it's pretty much guaranteed that the optimal ratio is different from 55/45. That means there is a different asset allocation that has lower risk and the same expected total portfolio return. This can clearly be observed in the third image in this thread.

[bobwuzbob]

I don't think a permanent position in VXX is the best way to improve risk-adjusted returns. It takes more than it gives over the long run. Now, if you could figure out how to 1) hold it strategically and 2) size the position appropriately, you'd be onto something.

If the goal is to improve risk-adjusted returns, the following strategy is effective and low-effort.

I agree about VXX, but keep in mind XVZ has a lot less drawdown. The lowest it ever fell from its initial value when it launched in Sept 2011 was 32%. I bring it up because adding even 5% to the portfolio decreases maximum drawdown at the cost of lowered returns. Here's a backtest showing what I'm talking about. VIX positively correlates with having better worst years, smaller maximum drawdowns, and worse total returns. It's a tradeoff that the investor can decide if they want to make. IMO though, the optimal VIX allocation will probably be small, but nonzero.

And if you use the portfolio optimizer to minimize maximum drawdown over the same time interval, it will actually put 1/3 of your assets in XVZ.

Also, that strategy looks interesting, but just to be clear, that's an active portfolio right?

I would recommend against putting anything in XVZ currently since volume is incredibly low. It has an average volume of just a bit more than 3,000. It looks fine in PV since it doesn't take into account how illiquid and low volume it is. If you plan on rebalancing regularly with M1 you will get terrible fills with this ETF.

[cos]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

Let me try my hand at those assumptions. The 55/45 UPRO/TMF portfolio happens to approximately maximize CAGR (and thus the Kelly Criterion) when simulating UPRO and TMF all the way back to January, 1987. My guess at the assumptions being made:

1. The investor necessarily has a CRRA of γ ≈ 1

2. No other assets meaningfully exist (i.e. lack of faith in factors)

3. Historical returns, volatility, and correlations will remain similar in the future

Am I wrong about any of these? What am I missing?

[Tingting1013]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

One of the assumptions is allocating 100% of your portfolio to this strategy.

If you bucketize HFEA away from your main portfolio, it's pretty much guaranteed that the optimal ratio is different from 55/45. That means there is a different asset allocation that has lower risk and the same expected total portfolio return. This can clearly be observed in the third image in this thread.

If you bucketize HFEA then there is no such thing as an “optimal total portfolio”. That is the point of bucketing, it is entirely separate from your core portfolio.

In which case the optimal AA of HFEA can be evaluated on its own without regard to any other assets.

[cos]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

One of the assumptions is allocating 100% of your portfolio to this strategy.

If you bucketize HFEA away from your main portfolio, it's pretty much guaranteed that the optimal ratio is different from 55/45. That means there is a different asset allocation that has lower risk and the same expected total portfolio return. This can clearly be observed in the third image in this thread.

If you bucketize HFEA then there is no such thing as an “optimal total portfolio”. That is the point of bucketing, it is entirely separate from your core portfolio.

In which case the optimal AA of HFEA can be evaluated on its own without regard to any other assets.

Nope. Money is fungible, and we're dealing exclusively with highly liquid assets. At all times, your portfolio consists of all highly liquid assets in your possession. Mentally separating your assets might make you feel better in the immediate, but it doesn't change the math around your constant relative risk aversion and might cause more pain further down the road if you fail to properly manage your risk exposure. In short, "bucketizing" might be an easier means of managing risk, but it's certainly a far less effective means relative to the utility function optimization approach discussed in Uncorrelated's linked thread.

[NMBob]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

Let me try my hand at those assumptions. The 55/45 UPRO/TMF portfolio happens to approximately maximize CAGR (and thus the Kelly Criterion) when simulating UPRO and TMF all the way back to January, 1987. My guess at the assumptions being made:

1. The investor necessarily has a CRRA of γ ≈ 1

2. No other assets meaningfully exist (i.e. lack of faith in factors)

3. Historical returns, volatility, and correlations will remain similar in the future


"when simulating UPRO and TMF all the way back to January, 1987."

Am I wrong about any of these? What am I missing?

I would say a lot depends on semantics.

"when simulating UPRO and TMF all the way back to January, 1987."

I would argue that sounds like the original implementation of 40/60. A decision was later made to prioritize periods of similar interest rates to pick the asset allocation. The AA was changed to 55/45 based on 3 specific periods as I recall. Read the very end of the 1st thread and beginning of this thread if that is new news to readers. Also, that makes me wonder, when is this to be assessed again and what is the criteria to change the AA?

"The 55/45 UPRO/TMF portfolio happens to approximately maximize CAGR"

It looks in those 3 similar interest rate time periods that at least one option 60/40 has a higher CAGR. Someone else posted for the entire time period of like 1955-2018 that 70/30 has the highest cagr.

[Uncorrelated]

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

Let me try my hand at those assumptions. The 55/45 UPRO/TMF portfolio happens to approximately maximize CAGR (and thus the Kelly Criterion) when simulating UPRO and TMF all the way back to January, 1987. My guess at the assumptions being made:

1. The investor necessarily has a CRRA of γ ≈ 1

2. No other assets meaningfully exist (i.e. lack of faith in factors)

3. Historical returns, volatility, and correlations will remain similar in the future

Am I wrong about any of these? What am I missing?

That is correct. There is a fourth assumption that is missing. If the investor is using this strategy for the long term, then the investor must have a human capital and expected social security benefits of zero. Alternatively, the investor can rationalize HFEA in a lifecycle investing framework if he is currently in the leverage constrained phase.

I have similar criticisms with hedgefundie's strategy itself. Is it a valid strategy? Yes, there is a certain set of assumptions that result in 55/45 being the optimal solution. I would attribute that to sheer luck. If you aren't aware what those assumptions are, it's very unlikely this is the right strategy for you.

This criticism is only relevant for those who are using HFEA as a core holding in their portfolio, which is a very small fraction of participants.

One of the assumptions is allocating 100% of your portfolio to this strategy.

If you bucketize HFEA away from your main portfolio, it's pretty much guaranteed that the optimal ratio is different from 55/45. That means there is a different asset allocation that has lower risk and the same expected total portfolio return. This can clearly be observed in the third image in this thread.

If you bucketize HFEA then there is no such thing as an “optimal total portfolio”. That is the point of bucketing, it is entirely separate from your core portfolio.

In which case the optimal AA of HFEA can be evaluated on its own without regard to any other assets.

You invest to meet your financial goals. Meeting your financial goals is a function of the probability distribution of total portfolio outcomes. By bucketing your portfolio, the optimization processes is compromised and you are guaranteed not to select the portfolio that is best to meet your financial goals.

Bucketing can be seen as a trade-off between simplicity and efficiency. If that's the trade-off you want to make, that's okay. But that doesn't change the underlying math: meeting your financial goals is still a function of your total portfolio. Some models even include your human capital and future social security benefits as part of your portfolio, this can result in large certainty equivalent (risk adjusted) gains.