HEDGEFUNDIE's excellent adventure Part II: The next journey

Discuss all general (i.e. non-personal) investing questions and issues, investing news, and theory.
perfectuncertainty
Posts: 126
Joined: Sun Feb 04, 2018 7:44 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by perfectuncertainty »

haranoth wrote: Wed Sep 09, 2020 2:50 pm
perfectuncertainty wrote: Wed Sep 09, 2020 1:20 pm
Mickelous wrote: Wed Sep 09, 2020 8:09 am
perfectuncertainty wrote: Tue Sep 08, 2020 11:01 pm For those of you wanting a 6-month volatility version for AAA here is an updated version.

I also added an additional AAA Inverse-Volatility Calculation that allows one to combine UPRO, TQQQ and UPRO and allocate accordingly. To duplicate this in PV here are the parameters and the backtest

AAA-6-Mo

Again, I prefer the 21-day version for reasons I've already stated.

It's showing the calculations through September 8th. PV doesn't have them updated yet, but they should match for a 6 month Volatility Period as the input, varying Inverse volatility, and Minimum Variance.
Not allowing me to access it now.
UPDATED Spreadsheet
Hi,
Is this google sheets automatically updated everyday, I couldn't find the place, where you pull the data automatically for UPRO, TMF e.g.
=GOOGLEFINANCE("NYSEARCA:UPRO","price",TODAY()-21,TODAY())

Just asking, since I would like to actually take your sheet and modify it to figure out how much of a difference AAA minimum variance vs AAA inverse volatility vs Will's 25% target volatility vs. HFEA original 55/45 portfolio makes :)
Thanks for providing the google sheet, it's good to see some of the associated equations and try to learn some of this lingo.
I didn't set it up to refresh each day. If folks want that feature I or someone else could do so.
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

jarjarM wrote: Wed Sep 09, 2020 1:31 pm
Here is the original AAA strategy proposed by smartly based on the white paper:
Strategy rules tested:
- At the close on the last trading day of the month, calculate the 6-month (126-day) return for each of the following 10 asset classes: US equities (represented by SPY), European equities (EZU), Japanese equities (EWJ), emerging market equities (EEM), US REITs (VNQ), international REITs (RWX), US 10-year Treasuries (IEF), US 30-year Treasuries (TLT), commodities (DBC) and gold (GLD).

- Go long at the close the five assets (i.e. half of the portfolio) with the highest 6-month return. Weight the five assets according to minimum variance optimization, using a “weighted” covariance matrix calculated based on 126-day correlation and 20-day volatility.

- Hold positions until the final trading day of the following month. Rebalance the entire portfolio monthly, regardless of whether there is a change in position.
Momentum (6 months look back) is build into minimum variance but not really applicable here. The original paper uses 10 assets class and pick the top 5. However, the covariance matrix is utilizing 6month look back (6*21 trading days). For momentum, it's picking the top 2 and go 50/50 (or 33.3/33.3/33.3 for top 3). Combo is combining momentum and inverse volatility together. Hope this helps. :beer
That is super helpful, thank you! As a follow up, how is the Combo combined? averaging the two weights?

For your Minimum Variance on this chart did you use the 126 day correlation and 20 day volatility? (i.e. does the new spreadsheet that kerstverlitchting so kindly provided include enough data points to fully be minimum variance (if it is only using 21 days data))?
occambogle
Posts: 526
Joined: Thu Dec 12, 2019 4:58 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by occambogle »

perfectuncertainty wrote: Wed Sep 09, 2020 2:55 pm I didn't set it up to refresh each day. If folks want that feature I or someone else could do so.
If it’s not a lot of work that would be useful to some I think, including those on Mac for whom the other spreadsheet doesn’t update.
User avatar
Meaty
Posts: 796
Joined: Mon Jul 22, 2013 7:35 pm
Location: Texas

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Meaty »

Has there been resolution on the inclusion of VXX? Something like 70 UPRO 25 TMF 5 VXX?
"Discipline equals Freedom" - Jocko Willink
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

Meaty wrote: Wed Sep 09, 2020 5:48 pm Has there been resolution on the inclusion of VXX? Something like 70 UPRO 25 TMF 5 VXX?
Adding VXX has being pretty bad to the performance of the portfolio since 2010 - 9/8/2020

Code: Select all

		UPRO only		70/25/5 Portfolio
cagr		28.96			28.72
sharpe		0.76			1.07
volatility	0.52			0.27
langlands
Posts: 569
Joined: Wed Apr 03, 2019 10:05 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by langlands »

jarjarM wrote: Wed Sep 09, 2020 6:23 pm
Meaty wrote: Wed Sep 09, 2020 5:48 pm Has there been resolution on the inclusion of VXX? Something like 70 UPRO 25 TMF 5 VXX?
Adding VXX has being pretty bad to the performance of the portfolio since 2010 - 9/8/2020

Code: Select all

		UPRO only		70/25/5 Portfolio
cagr		28.96			28.72
sharpe		0.76			1.07
volatility	0.52			0.27
Wait what? 70/25/5 looks like an improvement. CAGR is nearly identical, sharpe is up a lot, and volatility is down a lot.
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

langlands wrote: Wed Sep 09, 2020 6:46 pm
jarjarM wrote: Wed Sep 09, 2020 6:23 pm
Meaty wrote: Wed Sep 09, 2020 5:48 pm Has there been resolution on the inclusion of VXX? Something like 70 UPRO 25 TMF 5 VXX?
Adding VXX has being pretty bad to the performance of the portfolio since 2010 - 9/8/2020

Code: Select all

		UPRO only		70/25/5 Portfolio
cagr		28.96			28.72
sharpe		0.76			1.07
volatility	0.52			0.27
Wait what? 70/25/5 looks like an improvement. CAGR is nearly identical, sharpe is up a lot, and volatility is down a lot.
Sorry, this is what happens when I try to post something in between meeting in a hurry, I mean to compare to the OG 55/45 UPRO/TMF and its several variance.

Adding VXX has being pretty bad to the performance of the portfolio since 2010 - 9/8/2020

Code: Select all

		UPRO only		70/25/5 Portfolio	55/45OG-Portfolio	Inverse-Vol	AdapAssetAlloc
cagr		28.96			28.72				31.3			32.9		33.6
sharpe		0.76			1.07				1.24			1.31		1.31
volatility	0.52			0.27				0.24			0.24		0.25
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

RovenSkyfall wrote: Wed Sep 09, 2020 2:55 pm
jarjarM wrote: Wed Sep 09, 2020 1:31 pm
Here is the original AAA strategy proposed by smartly based on the white paper:
Strategy rules tested:
- At the close on the last trading day of the month, calculate the 6-month (126-day) return for each of the following 10 asset classes: US equities (represented by SPY), European equities (EZU), Japanese equities (EWJ), emerging market equities (EEM), US REITs (VNQ), international REITs (RWX), US 10-year Treasuries (IEF), US 30-year Treasuries (TLT), commodities (DBC) and gold (GLD).

- Go long at the close the five assets (i.e. half of the portfolio) with the highest 6-month return. Weight the five assets according to minimum variance optimization, using a “weighted” covariance matrix calculated based on 126-day correlation and 20-day volatility.

- Hold positions until the final trading day of the following month. Rebalance the entire portfolio monthly, regardless of whether there is a change in position.
Momentum (6 months look back) is build into minimum variance but not really applicable here. The original paper uses 10 assets class and pick the top 5. However, the covariance matrix is utilizing 6month look back (6*21 trading days). For momentum, it's picking the top 2 and go 50/50 (or 33.3/33.3/33.3 for top 3). Combo is combining momentum and inverse volatility together. Hope this helps. :beer
That is super helpful, thank you! As a follow up, how is the Combo combined? averaging the two weights?

For your Minimum Variance on this chart did you use the 126 day correlation and 20 day volatility? (i.e. does the new spreadsheet that kerstverlitchting so kindly provided include enough data points to fully be minimum variance (if it is only using 21 days data))?
Yes, the original backtest shown was using the same parameters as presented in smartly's strategy paper. I don't think the new spreadsheet uses 126 day correlation due to the fact that yahoo only allows download data of 100 days. I uses Tiingo API and python to do my backtest as my little side project during the pandemic, just for fun.
User avatar
cos
Posts: 223
Joined: Fri Aug 23, 2019 7:34 pm
Location: Boston
Contact:

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by cos »

jarjarM wrote: Wed Sep 09, 2020 7:42 pm Yes, the original backtest shown was using the same parameters as presented in smartly's strategy paper. I don't think the new spreadsheet uses 126 day correlation due to the fact that yahoo only allows download data of 100 days. I uses Tiingo API and python to do my backtest as my little side project during the pandemic, just for fun.
Have you shared the code elsewhere in the thread? If not, I'd love to see it!
User avatar
kerstverlichting
Posts: 27
Joined: Tue Feb 11, 2020 5:26 am
Contact:

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by kerstverlichting »

jarjarM wrote: Wed Sep 09, 2020 7:42 pm
RovenSkyfall wrote: Wed Sep 09, 2020 2:55 pm
jarjarM wrote: Wed Sep 09, 2020 1:31 pm
Here is the original AAA strategy proposed by smartly based on the white paper:
Strategy rules tested:
- At the close on the last trading day of the month, calculate the 6-month (126-day) return for each of the following 10 asset classes: US equities (represented by SPY), European equities (EZU), Japanese equities (EWJ), emerging market equities (EEM), US REITs (VNQ), international REITs (RWX), US 10-year Treasuries (IEF), US 30-year Treasuries (TLT), commodities (DBC) and gold (GLD).

- Go long at the close the five assets (i.e. half of the portfolio) with the highest 6-month return. Weight the five assets according to minimum variance optimization, using a “weighted” covariance matrix calculated based on 126-day correlation and 20-day volatility.

- Hold positions until the final trading day of the following month. Rebalance the entire portfolio monthly, regardless of whether there is a change in position.
Momentum (6 months look back) is build into minimum variance but not really applicable here. The original paper uses 10 assets class and pick the top 5. However, the covariance matrix is utilizing 6month look back (6*21 trading days). For momentum, it's picking the top 2 and go 50/50 (or 33.3/33.3/33.3 for top 3). Combo is combining momentum and inverse volatility together. Hope this helps. :beer
That is super helpful, thank you! As a follow up, how is the Combo combined? averaging the two weights?

For your Minimum Variance on this chart did you use the 126 day correlation and 20 day volatility? (i.e. does the new spreadsheet that kerstverlitchting so kindly provided include enough data points to fully be minimum variance (if it is only using 21 days data))?
Yes, the original backtest shown was using the same parameters as presented in smartly's strategy paper. I don't think the new spreadsheet uses 126 day correlation due to the fact that yahoo only allows download data of 100 days. I uses Tiingo API and python to do my backtest as my little side project during the pandemic, just for fun.
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
occambogle
Posts: 526
Joined: Thu Dec 12, 2019 4:58 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by occambogle »

perfectuncertainty wrote: Wed Sep 09, 2020 2:55 pm I also added an additional AAA Inverse-Volatility Calculation that allows one to combine UPRO, TQQQ and UPRO and allocate accordingly. To duplicate this in PV here are the parameters and the backtest
Is there a way to do a side-by-side comparison of these adaptive allocation models vs fixed allocation models (e.g. 55/45 UPRO/TMF, 52/48 TQQQ/TMF) in PV? I can't seem to work out how to do that....
perfectuncertainty
Posts: 126
Joined: Sun Feb 04, 2018 7:44 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by perfectuncertainty »

occambogle wrote: Thu Sep 10, 2020 1:41 am

Is there a way to do a side-by-side comparison of these adaptive allocation models vs fixed allocation models (e.g. 55/45 UPRO/TMF, 52/48 TQQQ/TMF) in PV? I can't seem to work out how to do that....
Not sure I understand what you are looking for, but I haven't seen side-by-side comparisons. I've run the backtests separately in PV and noted the results which include the regular outputs from PV. For AAA, you rebalance every month, whereas for fixed you set up the rebalance paramters.
User avatar
kerstverlichting
Posts: 27
Joined: Tue Feb 11, 2020 5:26 am
Contact:

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by kerstverlichting »

kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

EDIT: New version: viewtopic.php?p=5496322#p5496322

@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.
Last edited by kerstverlichting on Tue Sep 15, 2020 9:43 am, edited 2 times in total.
AnilG
Posts: 63
Joined: Wed Sep 09, 2015 1:41 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by AnilG »

20 day volatility = Standard Deviation of daily returns for past 20 trading days.

inverse volatility = 1 / volatility = 1/ Standard Deviation

Rebalancing using inverse volatility = portfolio allocation of an ETF = inverse volatility of the ETF / sum of inverse volatility of both ETFs.

Refer to Section 3.1 of following research paper. I used the equation in this section to calculate portfolio allocation.

Mailard S., Roncalli T., Teiletche J., On the properties of equally-weighted risk contributions portfolios, (May 2009)
http://www.thierry-roncalli.com/download/erc.pdf. Older version available at SSRN at https://papers.ssrn.com/sol3/papers.cfm ... id=1271972
worthit wrote: Wed Sep 09, 2020 10:17 am
AnilG wrote: Tue Sep 08, 2020 5:00 am I am using TQQQ/TMF strategy with monthly rebalancing using inverse volatility with 20 day volatility.
Now this is what I am not clear about. What does "rebalancing using inverse volatility with 20 day volatility" mean in the above response?

TIA.
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
nehawk87
Posts: 15
Joined: Sun Jun 30, 2019 8:49 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by nehawk87 »

kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?

Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image
Great stuff, really enjoying reading up on AAA and how to employ this strategy. I want to make sure I'm understanding this correctly.

I'm a little surprised at the AAA recommendation based on the data of 19.8% stock and 80.2% bond inverse volatility allocation compared to the typical 55/45 strategy. Is it common for AAA to lean heavier into the bond allocation? Curious what the experience has been for those who have been using AAA awhile and also want to ensure I'm interpreting this data correctly.
tomphilly
Posts: 59
Joined: Wed Aug 05, 2020 11:29 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by tomphilly »

It seems like the discussion has swung from a conversation around a higher stock to bond ratio (80/20) to the opposite. As I understand it, holding bonds in excess of 45% runs contrary to the risk profile of this entire strategy and its offshoots.

Maybe Uncorrelated can weigh in on this - it might be better to just deleverage than to run in reverse. 20/80 UPRO/TMF seems like a septuagenarian retirement fund on meth.
Perfect Uncertainty
Posts: 12
Joined: Wed Jun 24, 2020 10:20 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Perfect Uncertainty »

tomphilly wrote: Thu Sep 10, 2020 9:35 am It seems like the discussion has swung from a conversation around a higher stock to bond ratio (80/20) to the opposite. As I understand it, holding bonds in excess of 45% runs contrary to the risk profile of this entire strategy and its offshoots.

Maybe Uncorrelated can weigh in on this - it might be better to just deleverage than to run in reverse. 20/80 UPRO/TMF seems like a septuagenarian retirement fund on meth.
You checked the results from the back tests?
User avatar
willthrill81
Posts: 20911
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by willthrill81 »

Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
I don't think I have the...nerves...to handle that. While my returns have been beyond excellent (45% in 15 months), I don't know that I could hold on if I was use a large portion of my portfolio. Even Hedgefundie didn't go for that.

I genuinely hope that it works out for you. :beer
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

willthrill81 wrote: Thu Sep 10, 2020 10:21 am
Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
I don't think I have the...nerves...to handle that. While my returns have been beyond excellent (45% in 15 months), I don't know that I could hold on if I was use a large portion of my portfolio. Even Hedgefundie didn't go for that.

I genuinely hope that it works out for you. :beer
I'm only 31 so I figure I can handle the volatility. Best case scenario I retire in 10 years worst case oh well I'm young.
User avatar
willthrill81
Posts: 20911
Joined: Thu Jan 26, 2017 3:17 pm
Location: USA

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by willthrill81 »

Mickelous wrote: Thu Sep 10, 2020 10:34 am
willthrill81 wrote: Thu Sep 10, 2020 10:21 am
Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am

Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
I don't think I have the...nerves...to handle that. While my returns have been beyond excellent (45% in 15 months), I don't know that I could hold on if I was use a large portion of my portfolio. Even Hedgefundie didn't go for that.

I genuinely hope that it works out for you. :beer
I'm only 31 so I figure I can handle the volatility. Best case scenario I retire in 10 years worst case oh well I'm young.
I would have heart palpitations watching my account, which wouldn't be nearly as large as many around here, swing by tens of thousands of dollars or more on a regular basis. But if you can avoid looking at it, maybe you'll be able to stay the course. The upside potential is undeniably large.
“It's a dangerous business, Frodo, going out your door. You step onto the road, and if you don't keep your feet, there's no knowing where you might be swept off to.” J.R.R. Tolkien,The Lord of the Rings
tomphilly
Posts: 59
Joined: Wed Aug 05, 2020 11:29 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by tomphilly »

Perfect Uncertainty wrote: Thu Sep 10, 2020 10:16 am You checked the results from the back tests?
Yes, I did - to do a fairer comparison to HFEA or TV you need to remove TQQQ from the backtest or add it to HFEA's and TV's backtest - TQQQ makes any 10 year backtest look amazing. As far as I can tell, an AA model with UPRO & TMF significantly underperforms vanilla HFEA and the riskier TV strategy, as far as CAGR. Note the Vanguard benchmark is consistent across all three backtests. Also note that TV can be further CAGR optimized in practice by not exceeding 45% TMF when being signaled to do so by the PV forward volatility signal.

AA (UPRO & TMF only) - PV:
Image

HFEA UPRO/TMF 55/45 - PV:
Image

TV UPRO/TMF 80/20 @ 35% volatility - PV:
Image
Last edited by tomphilly on Thu Sep 10, 2020 10:55 am, edited 2 times in total.
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

I tried adding UGL to that spreadsheet replacing GLD and it would reference the data in the overview sheet although I got the data to work in the database. Sad that UGLD is no more.
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

tomphilly wrote: Thu Sep 10, 2020 10:41 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 10:16 am You checked the results from the back tests?
Yes, I did - to do a fairer comparison to HFEA or TV you need to remove TQQQ from the backtest or add it to HFEA's and TV's backtest - TQQQ makes any 10 year backtest look impressive. As far as I tell, an AA model with UPRO & TMF significantly underperforms vanilla HFEA and the riskier TV strategy, as far as CAGR. Note the Vanguard benchmark is consistent across all three backtests. Also note that TV can be further CAGR optimized by not exceeding 45% TMF when being signaled to do so by the PV forward signal.

AA (UPRO & TMF only) - PV:
Image

HFEA UPRO/TMF 55/45 - PV:
Image

TV UPRO/TMF 80/20 @ 35% volatility - PV:
Image
Here is a PV link here showing the difference between equally weighted (almost HFEA) vs AAA. If you go to "Timing Periods" you can see the breakdown of the different allocations for the AAA.

If your link is what you applied, I think you may have applied some incorrect features to the AA. From my understanding it ought to be 21 day volatility period and minimum variance for allocation weight.
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

willthrill81 wrote: Thu Sep 10, 2020 10:38 am
Mickelous wrote: Thu Sep 10, 2020 10:34 am
willthrill81 wrote: Thu Sep 10, 2020 10:21 am
Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am

kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
I don't think I have the...nerves...to handle that. While my returns have been beyond excellent (45% in 15 months), I don't know that I could hold on if I was use a large portion of my portfolio. Even Hedgefundie didn't go for that.

I genuinely hope that it works out for you. :beer
I'm only 31 so I figure I can handle the volatility. Best case scenario I retire in 10 years worst case oh well I'm young.
I would have heart palpitations watching my account, which wouldn't be nearly as large as many around here, swing by tens of thousands of dollars or more on a regular basis. But if you can avoid looking at it, maybe you'll be able to stay the course. The upside potential is undeniably large.
I just laugh on the down days like the last two times tqqq dropped 15 percent in a day recently. Just an opportunity to rebalance some tmf in there come rebalancing time. I take failure pretty well.
User avatar
kerstverlichting
Posts: 27
Joined: Tue Feb 11, 2020 5:26 am
Contact:

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by kerstverlichting »

Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
xxx

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
OK wow, I just put the thing together in two days in between work. 😅 please be careful as I didn't fully check everything and it heavily relies on perfect uncertainty's calculations (which I also didn't verify). Cool to hear folks are so happy with it btw. I only got a few k to put into this strategy myself but it's fun being part of a community to do it together with.
nehawk87 wrote: Thu Sep 10, 2020 8:31 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
Great stuff, really enjoying reading up on AAA and how to employ this strategy. I want to make sure I'm understanding this correctly.

I'm a little surprised at the AAA recommendation based on the data of 19.8% stock and 80.2% bond inverse volatility allocation compared to the typical 55/45 strategy. Is it common for AAA to lean heavier into the bond allocation? Curious what the experience has been for those who have been using AAA awhile and also want to ensure I'm interpreting this data correctly.
Ah yes, I checked this for you, seems to be the case because of the odd mix of TQQQ/UPRO/TMF/TYD in the preview. A "normal" UPRO/TMF allocation (21 days volatility) ends up with just under 60% in TMF:
Image

Seems more reasonable, right?

Still though, I advise people to double check everything and let me know if something seems amiss.
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

kerstverlichting wrote: Thu Sep 10, 2020 11:05 am
Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
xxx

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
OK wow, I just put the thing together in two days in between work. 😅 please be careful as I didn't fully check everything and it heavily relies on perfect uncertainty's calculations (which I also didn't verify). Cool to hear folks are so happy with it btw. I only got a few k to put into this strategy myself but it's fun being part of a community to do it together with.
nehawk87 wrote: Thu Sep 10, 2020 8:31 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
Great stuff, really enjoying reading up on AAA and how to employ this strategy. I want to make sure I'm understanding this correctly.

I'm a little surprised at the AAA recommendation based on the data of 19.8% stock and 80.2% bond inverse volatility allocation compared to the typical 55/45 strategy. Is it common for AAA to lean heavier into the bond allocation? Curious what the experience has been for those who have been using AAA awhile and also want to ensure I'm interpreting this data correctly.
Ah yes, I checked this for you, seems to be the case because of the odd mix of TQQQ/UPRO/TMF/TYD in the preview. A "normal" UPRO/TMF allocation (21 days volatility) ends up with just under 60% in TMF:
Image

Seems more reasonable, right?

Still though, I advise people to double check everything and let me know if something seems amiss.
I've been in a modified HFEA for 4 months now so this is just a tweak.
Perfect Uncertainty
Posts: 12
Joined: Wed Jun 24, 2020 10:20 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Perfect Uncertainty »

:moneybag
tomphilly wrote: Thu Sep 10, 2020 10:41 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 10:16 am You checked the results from the back tests?
Yes, I did - to do a fairer comparison to HFEA or TV you need to remove TQQQ from the backtest or add it to HFEA's and TV's backtest - TQQQ makes any 10 year backtest look amazing. As far as I can tell, an AA model with UPRO & TMF significantly underperforms vanilla HFEA and the riskier TV strategy, as far as CAGR. Note the Vanguard benchmark is consistent across all three backtests. Also note that TV can be further CAGR optimized in practice by not exceeding 45% TMF when being signaled to do so by the PV forward volatility signal.

AA (UPRO & TMF only) - PV:
Image

HFEA UPRO/TMF 55/45 - PV:
Image

TV UPRO/TMF 80/20 @ 35% volatility - PV:
Image
I use QLD rather than TQQQ or UPRO link
Last edited by Perfect Uncertainty on Thu Sep 10, 2020 11:24 am, edited 1 time in total.
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

RovenSkyfall wrote: Thu Sep 10, 2020 10:56 am
tomphilly wrote: Thu Sep 10, 2020 10:41 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 10:16 am You checked the results from the back tests?
Yes, I did - to do a fairer comparison to HFEA or TV you need to remove TQQQ from the backtest or add it to HFEA's and TV's backtest - TQQQ makes any 10 year backtest look impressive. As far as I tell, an AA model with UPRO & TMF significantly underperforms vanilla HFEA and the riskier TV strategy, as far as CAGR. Note the Vanguard benchmark is consistent across all three backtests. Also note that TV can be further CAGR optimized by not exceeding 45% TMF when being signaled to do so by the PV forward signal.

AA (UPRO & TMF only) - PV:
Image

HFEA UPRO/TMF 55/45 - PV:
Image

TV UPRO/TMF 80/20 @ 35% volatility - PV:
Image
Here is a PV link here showing the difference between equally weighted (almost HFEA) vs AAA. If you go to "Timing Periods" you can see the breakdown of the different allocations for the AAA.

If your link is what you applied, I think you may have applied some incorrect features to the AA. From my understanding it ought to be 21 day volatility period and minimum variance for allocation weight.
Tomphilly's original AA is using volatility of 6 months. If one uses 21 day volatility window, CAGR is ~4-5% better.

https://www.portfoliovisualizer.com/tes ... odWeight=0
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

cos wrote: Wed Sep 09, 2020 9:33 pm
jarjarM wrote: Wed Sep 09, 2020 7:42 pm Yes, the original backtest shown was using the same parameters as presented in smartly's strategy paper. I don't think the new spreadsheet uses 126 day correlation due to the fact that yahoo only allows download data of 100 days. I uses Tiingo API and python to do my backtest as my little side project during the pandemic, just for fun.
Have you shared the code elsewhere in the thread? If not, I'd love to see it!
Not yet, was going to fix a few items and publish it but got sidetracked by work project. Probably will jump back into fixing this little pet project in the next week or so I hope.
User avatar
kerstverlichting
Posts: 27
Joined: Tue Feb 11, 2020 5:26 am
Contact:

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by kerstverlichting »

Mickelous wrote: Thu Sep 10, 2020 11:11 am
kerstverlichting wrote: Thu Sep 10, 2020 11:05 amx.
I've been in a modified HFEA for 4 months now so this is just a tweak.
Nice, how is it doing thus far?
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

kerstverlichting wrote: Thu Sep 10, 2020 1:27 am
jarjarM wrote: Wed Sep 09, 2020 7:42 pm
RovenSkyfall wrote: Wed Sep 09, 2020 2:55 pm
jarjarM wrote: Wed Sep 09, 2020 1:31 pm
Here is the original AAA strategy proposed by smartly based on the white paper:
Strategy rules tested:
- At the close on the last trading day of the month, calculate the 6-month (126-day) return for each of the following 10 asset classes: US equities (represented by SPY), European equities (EZU), Japanese equities (EWJ), emerging market equities (EEM), US REITs (VNQ), international REITs (RWX), US 10-year Treasuries (IEF), US 30-year Treasuries (TLT), commodities (DBC) and gold (GLD).

- Go long at the close the five assets (i.e. half of the portfolio) with the highest 6-month return. Weight the five assets according to minimum variance optimization, using a “weighted” covariance matrix calculated based on 126-day correlation and 20-day volatility.

- Hold positions until the final trading day of the following month. Rebalance the entire portfolio monthly, regardless of whether there is a change in position.
Momentum (6 months look back) is build into minimum variance but not really applicable here. The original paper uses 10 assets class and pick the top 5. However, the covariance matrix is utilizing 6month look back (6*21 trading days). For momentum, it's picking the top 2 and go 50/50 (or 33.3/33.3/33.3 for top 3). Combo is combining momentum and inverse volatility together. Hope this helps. :beer
That is super helpful, thank you! As a follow up, how is the Combo combined? averaging the two weights?

For your Minimum Variance on this chart did you use the 126 day correlation and 20 day volatility? (i.e. does the new spreadsheet that kerstverlitchting so kindly provided include enough data points to fully be minimum variance (if it is only using 21 days data))?
Yes, the original backtest shown was using the same parameters as presented in smartly's strategy paper. I don't think the new spreadsheet uses 126 day correlation due to the fact that yahoo only allows download data of 100 days. I uses Tiingo API and python to do my backtest as my little side project during the pandemic, just for fun.
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
I'll take a look at it tonight. Though I think 100 days or 126 days may not make much of a difference and any positive or negative will be more due to overfitting of past performance data. However, one thing I would say though is that there's real evidence (either thru backtest or academic study) that trading day makes difference. Last trading of the month is generally use for the calculation and then trade takes place using the first day of trading in next month. But one shifts to the trading day to middle of the month (11 day offset), one loses 2% in CAGR, or if one shift trading day toward 2nd 1/2 of the month (15 day offset), one loses 5% CAGR. Very crazy. I'll try to pose some of these backtest nugget tonight. But everyone should seriously consider their rebalance day, just food for thought. :beer
tomphilly
Posts: 59
Joined: Wed Aug 05, 2020 11:29 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by tomphilly »

jarjarM wrote: Thu Sep 10, 2020 11:15 am Tomphilly's original AA is using volatility of 6 months. If one uses 21 day volatility window, CAGR is ~4-5% better.
I just used a previously provided backtest by perfectuncertainty and removed TQQQ for a fairer comparison:
perfectuncertainty wrote: Wed Sep 09, 2020 2:55 pm I also added an additional AAA Inverse-Volatility Calculation that allows one to combine UPRO, TQQQ and UPRO and allocate accordingly. To duplicate this in PV here are the parameters and the backtest
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

kerstverlichting wrote: Thu Sep 10, 2020 11:19 am
Mickelous wrote: Thu Sep 10, 2020 11:11 am
kerstverlichting wrote: Thu Sep 10, 2020 11:05 amx.
I've been in a modified HFEA for 4 months now so this is just a tweak.
Nice, how is it doing thus far?
After the drawdown the past week I'm up 20 percent.
Perfect Uncertainty
Posts: 12
Joined: Wed Jun 24, 2020 10:20 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Perfect Uncertainty »

I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

Perfect Uncertainty wrote: Thu Sep 10, 2020 11:13 am

I use QLD rather than TQQQ or UPRO link
It looks like if you use minimum variance you might do a little bit better according to the PV link you sent.
tomphilly
Posts: 59
Joined: Wed Aug 05, 2020 11:29 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by tomphilly »

Perfect Uncertainty wrote: Thu Sep 10, 2020 11:32 am I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
Do you have an AA backtest with comparable CAGR to this 80/20 35% vol TV model?
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

jarjarM wrote: Thu Sep 10, 2020 11:23 am
I'll take a look at it tonight. Though I think 100 days or 126 days may not make much of a difference and any positive or negative will be more due to overfitting of past performance data. However, one thing I would say though is that there's real evidence (either thru backtest or academic study) that trading day makes difference. Last trading of the month is generally use for the calculation and then trade takes place using the first day of trading in next month. But one shifts to the trading day to middle of the month (11 day offset), one loses 2% in CAGR, or if one shift trading day toward 2nd 1/2 of the month (15 day offset), one loses 5% CAGR. Very crazy. I'll try to pose some of these backtest nugget tonight. But everyone should seriously consider their rebalance day, just food for thought. :beer
Thank you so much for pointing this out. I was literally wondering that this morning...
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

Perfect Uncertainty wrote: Thu Sep 10, 2020 11:32 am I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
Thank you PU, can you further explain this? I am trying to understand this topic as much as possible. It looked like for the original AAA they used minimum variance with a 6 month correlation and 21 day volatility. Wouldnt 6 months of correlation give one a better idea of actual correlation of asset classes rather than noise?
jarjarM
Posts: 276
Joined: Mon Jul 16, 2018 1:21 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by jarjarM »

RovenSkyfall wrote: Thu Sep 10, 2020 11:56 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 11:32 am I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
Thank you PU, can you further explain this? I am trying to understand this topic as much as possible. It looked like for the original AAA they used minimum variance with a 6 month correlation and 21 day volatility. Wouldn't 6 months of correlation give one a better idea of actual correlation of asset classes rather than noise?
The original AAA is based on 10 assets classes with far lower volatility than current LETFs due to leverage. Upon further examination of the original white paper and then looking at the LETFs behavior (distribution and etc), I think PU is on point in terms of using 1 month volatility (21 trading days). As for period for correlation, I think due to larger movement of the LETFs compare to original assets, significant correlation can be established in 1 month as well (remember correlation is heavily driven by outlier heavily leveraged point in statistical sense).
perfectuncertainty
Posts: 126
Joined: Sun Feb 04, 2018 7:44 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by perfectuncertainty »

RovenSkyfall wrote: Thu Sep 10, 2020 11:50 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 11:13 am

I use QLD rather than TQQQ or UPRO link
It looks like if you use minimum variance you might do a little bit better according to the PV link you sent.
MV does have the best CAGR result - also, note that there is no losing year (the worst year was +8.86%).

I use TQQQ, QLD, QQQ, TMF and TLT and put it into Inverse Volatility. Deleveraging with QQQ and TLT suits me.
Last edited by perfectuncertainty on Thu Sep 10, 2020 2:03 pm, edited 1 time in total.
perfectuncertainty
Posts: 126
Joined: Sun Feb 04, 2018 7:44 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by perfectuncertainty »

tomphilly wrote: Thu Sep 10, 2020 11:50 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 11:32 am I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
Do you have an AA backtest with comparable CAGR to this 80/20 35% vol TV model?
Try this one: QLD Minimum Variance.

The CAGR lags by 2%, however, look at the risk adjusted return. No losing years, Lower SD, Much smaller DD, Higher Sharpe and Higher Sortino. As a risk adjusted return I much prefer it.
perfectuncertainty
Posts: 126
Joined: Sun Feb 04, 2018 7:44 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by perfectuncertainty »

jarjarM wrote: Thu Sep 10, 2020 12:26 pm
RovenSkyfall wrote: Thu Sep 10, 2020 11:56 am
Perfect Uncertainty wrote: Thu Sep 10, 2020 11:32 am I put that link up because others wanted 6 months of volatility.

I don’t agree with 6 months of volatility or correlation coefficient usage for AAA. The nature of AAA is to rebalance every month so keeping 21 days is my recommendation. Otherwise set the trading frequency to 6 months too.
Thank you PU, can you further explain this? I am trying to understand this topic as much as possible. It looked like for the original AAA they used minimum variance with a 6 month correlation and 21 day volatility. Wouldn't 6 months of correlation give one a better idea of actual correlation of asset classes rather than noise?
The original AAA is based on 10 assets classes with far lower volatility than current LETFs due to leverage. Upon further examination of the original white paper and then looking at the LETFs behavior (distribution and etc), I think PU is on point in terms of using 1 month volatility (21 trading days). As for period for correlation, I think due to larger movement of the LETFs compare to original assets, significant correlation can be established in 1 month as well (remember correlation is heavily driven by outlier heavily leveraged point in statistical sense).
Spot On JarJarM. The rebalancing ideally occurs monthly in a tax-advantaged account. With monthly rebalancing, we are adjusting to "current" conditions. With the spreadsheets that I and others have put out there - you can see daily changes in the recommended asset allocation. I would argue that one might be best served to use a similar approach to that of traditional banded rebalancing. Rebalance each month and if the asset allocation for the realtime or daily AAA model moves beyond certain thresholds (like 8%) consider rebalancing sooner. You might not outperform slower rebalancing, but from my testing, it's pretty close and or better and it works itself out quite nicely. I use the real-time model because it addresses slippage that occurs between EOD models and gaps in openings to avoid chasing your tail.
RovenSkyfall
Posts: 82
Joined: Wed Apr 01, 2020 11:40 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by RovenSkyfall »

perfectuncertainty wrote: Thu Sep 10, 2020 2:11 pm
Spot On JarJarM. The rebalancing ideally occurs monthly in a tax-advantaged account. With monthly rebalancing, we are adjusting to "current" conditions. With the spreadsheets that I and others have put out there - you can see daily changes in the recommended asset allocation. I would argue that one might be best served to use a similar approach to that of traditional banded rebalancing. Rebalance each month and if the asset allocation for the realtime or daily AAA model moves beyond certain thresholds (like 8%) consider rebalancing sooner. You might not outperform slower rebalancing, but from my testing, it's pretty close and or better and it works itself out quite nicely. I use the real-time model because it addresses slippage that occurs between EOD models and gaps in openings to avoid chasing your tail.
So am I understanding correctly? You are suggesting AAA monthly rebalancing may be combined with typical banded rebalancing to further diminish losses or jump onto gains earlier than the monthly rebalancing?

Also, can you dumb down your last sentence?
I use the real-time model because it addresses slippage that occurs between EOD models and gaps in openings to avoid chasing your tail.
What are slippage, EOD models and gaps in openings. Thank you!
Perfect Uncertainty
Posts: 12
Joined: Wed Jun 24, 2020 10:20 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Perfect Uncertainty »

If you get a daily AAA reading and it says allocate 35% to UPRO and you want to execute the order the next day. Let’s say it’s a highly volatile period and UPRO gaps down over 5% at the open. I personally like to have the AAA close to real-time so I can execute my order accordingly
e5116
Posts: 600
Joined: Mon Oct 05, 2009 11:22 am

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by e5116 »

kerstverlichting wrote: Thu Sep 10, 2020 11:05 am
Mickelous wrote: Thu Sep 10, 2020 8:21 am
RovenSkyfall wrote: Thu Sep 10, 2020 8:13 am
kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
xxx

@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.
kerstverlichting thank you so much for working at this. It is really amazing that you are bringing people a tool to help optimize their investing.
Yeah no joke I have almost my entire 401k and Roth IRA in this strategy.
OK wow, I just put the thing together in two days in between work. 😅 please be careful as I didn't fully check everything and it heavily relies on perfect uncertainty's calculations (which I also didn't verify). Cool to hear folks are so happy with it btw. I only got a few k to put into this strategy myself but it's fun being part of a community to do it together with.
nehawk87 wrote: Thu Sep 10, 2020 8:31 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
Great stuff, really enjoying reading up on AAA and how to employ this strategy. I want to make sure I'm understanding this correctly.

I'm a little surprised at the AAA recommendation based on the data of 19.8% stock and 80.2% bond inverse volatility allocation compared to the typical 55/45 strategy. Is it common for AAA to lean heavier into the bond allocation? Curious what the experience has been for those who have been using AAA awhile and also want to ensure I'm interpreting this data correctly.
Ah yes, I checked this for you, seems to be the case because of the odd mix of TQQQ/UPRO/TMF/TYD in the preview. A "normal" UPRO/TMF allocation (21 days volatility) ends up with just under 60% in TMF:
Image

Seems more reasonable, right?

Still though, I advise people to double check everything and let me know if something seems amiss.
I must be dumb beacuse I'm missing something. I thought the spreadsheet defaults to 21 days volatility? What cell did you update to get the resulting more "normal" UPRO/TMF with under 60% in TMF. I, too, am getting like 22% UPRO/78% TMF. (Note: I'm not implementing this strategy...yet...just trying to learn). Thank you!
Mickelous
Posts: 132
Joined: Mon Apr 20, 2020 11:24 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Mickelous »

kerstverlichting wrote: Thu Sep 10, 2020 5:48 am
kerstverlichting wrote: Thu Sep 10, 2020 1:27 am [...]
My sheet indeed only goes up to 100 days. But also, it uses this range (that you can choose) for both correlation and volatility. From what I understand now, I should adjust it so that the correlation and volatility ranges can be defined separately. Eg. You could set the correlation range to 100 days (4.75mos) and the volatility range to 21 days (1mo), and then it'd work as required? (except of course the correlation range still wouldn't be 6mos due to the Yahoo restriction, but at least we got fairly close)

If you would be so kind, could you maybe run a back test with 100 days of correlation data and compare it to 126 days so that we can see if there is a significant difference (and whether it's positive or negative)?
Alright, I just updated my file (v3.0 now) to include a separate drop down to set the correlation range. This way, if you're interested in using Minimum Variance, you can set it to the max of 4.75mos/100 trading days (advised), and then set the volatility range separately (21 days advised) to get to the correct calculation of the Minimum Variance asset distribution. Note that it seems one would ideally use 6mos/126 trading days for the correlation range, but this is not possible with my file due to restrictions on Yahoo data that it uses.

Download: http://s000.tinyupload.com/index.php?fi ... 7640950076
Preview:
Image

@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.
Would you be able to fit 2x gold UGL in there? I wasn't able to get it working with the steps. Too bad the 3x gold etf got shut down, unless there is another one I'm not aware of.
NMBob
Posts: 362
Joined: Thu Apr 23, 2015 8:13 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by NMBob »

it looks like you can buy 3x ugld as an otc as ugldf
Impatience
Posts: 176
Joined: Thu Jul 23, 2020 3:15 pm

Re: HEDGEFUNDIE's excellent adventure Part II: The next journey

Post by Impatience »

The reason you guys are seeing such low UPRO/equity allocations is because of the recent volatility. The whole point of the adaptive asset allocation is that you’re rotating out of the asset with increasing volatility until it settles back down.

There’s a lot of strategies out there where you move out of stocks when the S&P breaks below a certain moving average, usually the 200 day. That movement is a proxy for high volatility. This is just a more rigorous way of doing that.

When the market moves sharply down as it has been doing the past week, the math will tell you to go heavy into bonds. This will probably reduce your drawdowns but could also limit your upside if we recover in a big way. However the idea is that volatility usually begets volatility and you’ll more likely gain by playing it safe until it decreases... probably.
Post Reply