Playing with momentum

[MustacheAndaHalf]

Next update will be on Dec 17, not Dec 15.

I'd like to reduce the time I spend checking on this experiment, so I've decided to revert back to updates every 5 weeks starting on Dec 17, 2018.  It's harder to keep motivated over an experiment that clearly lags the market.  So here's the rough rules of the experiment:

1. use the etfdb.com charts to find ETFs with the highest non-leveraged performance in the past 52 weeks
2. calculate momentum using performance data from a more trusted website (morningstar.com):
   52 week performance minus 4 week performance equals "one year momentum" (11 months, actually)
3. the fund switches 100% of assets into the U.S. equity ETF with the highest one year momentum
4. repeat every 5 weeks

Next update on Monday, December 17.

[MustacheAndaHalf]

Make that Dec 18th, a day late.

I took the top 6 entries on the top 100 list from etfdb, and skipped over a lots of ETFs that weren't 95%+ U.S. Equities.  Several ETNs (exchange traded note - a promise from a bank, no actual equities), various country ETFs, gas ETFs, etc.  Then I took the top 3 from that list, and compared performance using morningstar data:

#1) PSCH (again), 13.57 - (-10.72) = +24.29 mtm
#2) IHI (another health care ETF), 13.93 - (-8.70) = +22.63 mtm
#3) XWEB, 14.89 - (-5.59) = +20.48 mtm

So the fund remains invested 100% in PSCH.

Thanks to the recent market declines (including the -10.72% in 1 month shown above), the 5W fund is now back where it started.  It has a total profit of +0.55%, round that up to +0.6% for 2.8 yrs of investing.  It's benchmark S&P 500 is up +31.88% during the same time period.  Essentially, 10%/year versus 0%/year.

Next update on Jan 21, 2019.  Wishing your assets a happy new year...

[MustacheAndaHalf]

The U.S. equities markets won't open until Jan 22nd, so I'll place trades for this experiment at that time.  Until then, the relative positions of ETFs can't really change, so I can make my calculation in advance.  This time I'm going to show how I go about this in detail, so you see an example of the "algorithm" in action.

I visit the URL https://etfdb.com/compare/highest-52-week-returns/no-leveraged/ to see the current top 100 ETFs for the past 52 weeks.  I select only ETFs (not ETNs), without leverage, with 95%+ equities (not natural gas).  Here's the first ~30 from that list:

         
VIIX   VelocityShares Daily Long VIX Short-Term ETN   37.61%   not an ETF
VXX   iPath S&P 500 VIX Short-Term Futures ETN   37.54%   not an ETF
VIXY   ProShares VIX Short-Term Futures ETF   36.30%   not an ETF
VXXB   iPath Series B S&P 500 VIX Short-Term Futures ETN   35.95%   not an ETF
PALL   Aberdeen Standard Physical Palladium Shares ETF   24.24%   not U.S. equities
PSJ   Invesco Dynamic Software ETF   21.81%   roughly +4% momentum
XWEB   SPDR S&P Internet ETF   21.72%   roughly +3% momentum
QAT   iShares MSCI Qatar ETF   21.21%   not U.S. equities
KSA   iShares MSCI Saudi Arabia ETF   19.89%   not U.S. equities
VXZ   iPath S&P 500 VIX Mid-Term Futures ETN   18.98%   not an ETF
OLD   Long-Term Care ETF   18.88%   roughly +13% momentum
VXZB   iPath Series B S&P 500® VIX Mid-Term Futures ETN   18.70%   not an ETF
VIXM   ProShares VIX Mid-Term Futures ETF   18.08%   not U.S. equities
REZ   iShares Residential Real Estate ETF   17.70%   roughly +14% momentum
UNL   United States 12 Month Natural Gas Fund   16.85%   not U.S. equities
PUI   Invesco DWA Utilities Momentum ETF   16.34%   roughly +16% momentum
IHI   iShares U.S. Medical Devices ETF   14.31%   roughly +2% momentum
BTAL   AGFiQ US Market Neutral Anti-Beta Fund   14.11%   not U.S. equities
IGV   iShares North American Tech-Software ETF   13.86%   roughly -2% momentum
UNG   United States Natural Gas Fund   12.78%   not U.S. equities
XSW   SPDR S&P Software & Services ETF   12.75%   roughly -4% momentum
UTES   Reaves Utilities ETF   12.46%   roughly +13% momentum
FDN   First Trust Dow Jones Internet Index   12.36%   roughly -7% momentum
XITK   SPDR FactSet Innovative Technology ETF   12.09%   roughly -6% momentum
ARKG   ARK Genomic Revolution Multi-Sector ETF   12.07%   roughly -18% momentum
PSCH   Invesco S&P SmallCap Health Care ETF   11.51%   (current allocation)  roughly -4% momentum
CLIX   ProShares Long Online/Short Stores ETF   11.19%   leveraged + not 95% U.S. equities
JHMU   John Hancock Multi-Factor Utilities ETF   11.16%   roughly +12% momentum
GULF   WisdomTree Middle East Dividend Fund   11.15%   not U.S. equities
SKYY   First Trust ISE Cloud Computing Index Fund   11.13%   roughly -3% momentum
XHE   SPDR S&P Health Care Equipment ETF   11.09%   roughly -3% momentum
BFIT   Global X Health & Wellness Thematic ETF   11.01%   roughly +1% momentum
         

Ignoring all the ETNs, non-U.S. funds, non-equity funds... and then ignoring momentum below +10%, I get the following list of 5 ETFs:
PUI   Invesco DWA Utilities Momentum ETF   16.34%   roughly +16% momentum
REZ   iShares Residential Real Estate ETF   17.70%   roughly +14% momentum
OLD   Long-Term Care ETF   18.88%   roughly +13% momentum
UTES   Reaves Utilities ETF   12.46%   roughly +13% momentum
JHMU   John Hancock Multi-Factor Utilities ETF   11.16%   roughly +12% momentum


In the past I've spotted problems with etfdb data (# holdings, % cash, etc) so I check morningstar's data for each of these ETFs to get my accurate momentum data.  I will use the "performance" tab, and take 1 year minus 1 month performance:

PUI = 16.37 - (-.63) = +17 mtm
REZ = 17.76 - 2.00 = +15.76 mtm
OLD = 18.89 - 2.70 = +16.19 mtm
UTES = 12.54 - (-.57) = 13.11 mtm
JHMU = 12.21 - (-1.60) = 13.81 mtm

I could almost say "I'm getting OLD", but actually PUI has slightly higher momentum:
PUI is Invesco DWA Utilities Momentum ETF.  When the markets open Jan 22 (Tues), I'll shift the 5W fund experiment from 100% PSCH into 100% PUI - from healthcare momentum to utilities momentum.

[MustacheAndaHalf]

On Jan 22 I switched from 100% PSCH to 100% PUI (Investo DWA Utilities Momentum) per my prior update.  I forgot to mention the next update, which will be on Feb 22, 2019.

This experiment is turning 3 years old next month.  I've been using "Vanguard S&P 500 ETF" (VOO) as the benchmark.  According to my records, when I started this experiment (on Feb 17, 2016) VOO was trading at $178.22 / share.  It now trades at $242.13, for a share price growth of +35.86% (which fortunately matches my spreadsheet!).  In that same time, the momentum experiment has gained +4.48% overall.  The last quarter of 2018 did an immense amount of damage to performance, losing -28% in those 3 months, wiping out most of it's prior profit.  It's actually impressive how fast it fell, and perhaps a warning.

On an amusing note, last year I downsized the fund a bit.  The percentages are all the same, but I started a separate spreadsheet and copied a summary of prior performance into a new spreadsheet.  So the fund was much larger when it gained +24%, and much smaller when it lost -28%.  I couldn't have predicted that, but got lucky on the timing anyways.  The fund remains a very tiny allocation relative to my portfolio.

To break those losses down into a bit more detail:
from 2018-09-03 to 2018-10-25 the fund held PSCH (health care) for a -15% loss
from 2018-10-25 to 2018-11-12 the fund held DBO (oil) for a -8% loss
from 2018-11-12 to 2019-01-22 the fund returned to PSCH (health care) to grab another -7% loss
(the fund gained +15% while holding VGT for most of 2018)

Overall, especially owing to the 4th quarter of 2018, this momentum experiment still badly lags the S&P 500.

Next update on Feb 22.

[MustacheAndaHalf]

I said the next update is Feb 22, but my spreadsheet indicates I should make the purchase on Feb 25.  I think the best compromise is to commit to what I purchase today, Feb 22, but then make the actual purchase on Monday, Feb 25.

I used etfdb to narrow the list of top performing ETFs, skipping any non-U.S. non-stock ETFs.  That left me with 3 entries, with the following 1 year momentum (the latest month is subtracted to ignore reversion to the mean effects):

REZ) iShares Residential Real Estate Capped ETF, 30.61 - 4.63 = +26.0 mtm
OLD) The Long-Term Care ETF, 27.67 - 4.29 = + 23.4 mtm
PUI) Invesco DWA Utilities Momentum ETF, 25.82 - 5.64 = +20.2 mtm

So on Monday, Feb 25th, the experiment will switch from PUI (utilities) to REZ (real estate).

[MustacheAndaHalf]

Done.  Sold PUI when it was up +1% for the day, and REZ hadn't made any gains/losses.  The 5W fund experiment is now 100% invested in REZ (iShares Residential Real Estate ETF).  Something tells me switching investments every 5 weeks won't play out that well.

I wonder if there's a combination of checking often, but then not switching investments that often.  For example, after a purchase the fund could always wait 2-3 months.  And after the waiting period, check monthly.  At any rate, the experiment has no such provision right now.

Happy Birthday 5W fund experiment - now 3 years old.  Here's the performance during the entire experiment:
5W experiment: +10.3% total, or 3.3% / year for 3 years.
S&P 500 index: +44.9% total, or 13.1% / year for 3 years.
(sanity check: morningstar reports 15.1% / year for 3 years,
 and etfdb shows +54.6% return over 3 years)

Note this experiment doesn't track distributions / dividends.  Those would be annoying to track, and the dividends would only matter to the extent the S&P 500 had more or less than the fund experiment's holdings.  So that probably explains most of the gap between other data sources and my S&P 500 performance.  I use the GOOGLEFINANCE() function of Google Spreadsheets to track prices.

Back to the data, the experiment always sells during a year while a passive investor is assumed to hold the S&P 500 for years.  So a short-term tax rate of 22% is used for the experiment and a 15% tax rate is used for the benchmark S&P 500:
5W experiment) 10.3% total x 0.78 = +8.0% after tax
S&P 500 index) 44.9% total x 0.85 = +38.2% after tax
So the 5W experiment trails the S&P 500 by about 10%/year after taxes (and about 10%/year before taxes, too).

Next update is April 1, and hopefully the markets have a good joke in store for the experiment...

[sol]

I'm still reading along and following your progress.  The old momentum threads were one of the forum's first serious attempts to promote market timing, and we spilled thousands of words arguing about it.  Some die hard forum members remained convinced until their dying (banned) days that there was a persistent and exploitable signal in the historical stock market records that would tell them when to buy and when to sell assets of various types.

And the only way to really test that theory is to run the experiment, forward in time.  Data mining the historical records will always find something that looks promising, but that doesn't mean the same something will work in the future.  I applaud your dedication to accountability in this thread, and I hope other people will also find it instructive. 

Trying to time the market is hard work, and hardly ever seems to work out for a profit. 

[maginvizIZ]

It's been fun watching this unfold.

I bought MTUM (iShares momentum "smart etf") right around the time you started this thread. 2017 was a rip your face off good year (37.5% vs SPY = 21.83%).  So far YTD I'm lagging 1.5% from SPY.

I did just notice the turnover is 104%. :|

Anyway enough about that fund.  Again, cool to see you update this strategy.

[MustacheAndaHalf]

According to morningstar data, "iShares Edge MSCI USA Momentum Factor ETF" (MTUM) as of 2/28/2019 had beaten the S&P 500 for the past 3 years: 18%/year vs 15%/year.  Meanwhile, this experiment can't pick individual stocks - just entire ETFs with mixed momentum.  MTUM's calculations impress me, since they incorporate standard deviation into their filters (which would be very painful for me to re-create for my experiment, so I didn't).

I tried creating a second experiment on this forum where I picked individual micro-cap stocks (under 50M in assets).  I really, really should have considered standard deviation!  It was too painful watching 5-10% daily swings, and I gave up the experiment.  Although I kept invested through the 2008 crisis, I found my "stomach acid" limit for investing at that time.

I think it's unfortunate that I had to modify the rules of the experiment a few times.  Early on, I wanted to limit turnover and bumbled for a way to do that.  Later I realized I could pick any ETF not just sector ETFs, and then wound up picking from all ETFs since Vanguard began charging $0/trade for that.  Where in theory performance should have been even better, it didn't turn out that way.  I wish I hadn't bought the oil ETF, because it's such a bad fit for the experiment - but I also felt the rules can't be changed on the fly.  So I had to limit to U.S. equity ETFs later.

I think it could be interesting to see how a "Buffet small cap" experiment would go.  That would essentially test the Fama-French five factor model: small, value, quality, profitability (the 5th factor is beta, or the market itself).  And then track that against the S&P 500.

[arebelspy]

Just caught up on a bunch of posts in this thread.

I appreciate your willingness to continue updating us.

Most people advocating for something unusual stop updating once they aren't outperforming. Having years of data is interesting, even if it's not showing what you've hoped. So thank you.

I hope it turns around for you, even with just your play money.

[MustacheAndaHalf]

A year ago I moved the experiment from taxable to an IRA, and didn't mention that I reduced the amount of assets involved.  So I actually got luckier than the numbers show, but that's not part of the experiment.  I turned the prior performance of the experiment and it's benchmark into multipliers so that I could incorporate the prior 2 years, and still use a smaller allocation:
Feb 2016 - Feb 2018: +24.2% (larger allocation in taxable)
Feb 2018 - Feb 2019: -11.6% (smaller allocation in an IRA)

Another advantage of running an experiment like this: I can take my interest in experimenting and focus it here, instead of changing my core investments.  Much like a surprising number of money managers, this experiment is active, but privately I'm indexing.

Next update on April 1, 2019.

[Financial.Velociraptor]

@MustacheAndaHalf Where are you downloading your momentum numbers?

[MustacheAndaHalf]

So here's my new approach:
1. Every 2.5 weeks, visit the etfdb page for top 52 week returns (ignoring leveraged ETFs):
http://etfdb.com/compare/highest-52-week-returns/no-leveraged/
2. Check momentum of at least the top 5 ETFs.  Pick the ETF with the highest 52-week performance minus 4-week performance, and switch the fund's assets to that ETF.
3. If Vanguard charges a fee for that ETF, go down the list until a $0 trade can be found.

I switched to using etfdb in September 2018, and hand calculate momentum from their data.  I look up the "performance" tab and use 52-week minus 4-week performance to give me the last year less the most recent month.  Recently, I look at 10-25 entries by hand, and filter out those nowhere near the top few.  I then look up those top few on morningstar, which I think has more accurate data than etfdb (for example, incorect # of stocks on etfdb).  Both of those are free to use, which is another criteria for keeping the experiment's costs low.

[MustacheAndaHalf]

I'll calculate the highest momentum today, then purchase it tomorrow.

I visited "etfdb.com" which has 52-week performance charts.  I take the top ~10 entries, and look at etfdb.com performance (1 year minus 4 week).  Anything too far from the leaders gets dropped, leaving me with 3-4 ETFs.  This week I skipped an ETF named "MJ" because it's mostly foreign stocks.  Here's the "1 year minus 1 month" calculations of momentum (made by hand):

PSJ ("Invesco Dynamic Software ETF") : 31.70 - 3.78 = +27.92 mtm
ARKG ("ARK Genomic Revolution Multi-Sector ETF") : 30.27 - 6.82 = 23.45 mtm
IHI ("iShares U.S. Medical Devices ETF") : 25.46 - 2.36 (etfdb listed this as -1%) = 23.1 mtm

Those numbers were much closer using etfdb data, which is why I double check here.  The winner is PSJ ("dynamic software ETF").  On Monday I'll switch the 5W fund / experiment to 100% in PSJ.

When this experiment first started, the time it did well corresponded to a time when it kept the same holding (Vanguard Utilities).  Which sometimes make me wonder if turnover hurts the fund's performance (much as turnover hurts personal performance in studies).

[arebelspy]

When this experiment first started, the time it did well corresponded to a time when it kept the same holding (Vanguard Utilities).  Which sometimes make me wonder if turnover hurts the fund's performance (much as turnover hurts personal performance in studies).

So you're saying that buy and hold might be the key to not having momentum underperform?

[MustacheAndaHalf]

While I should be a cheerleader for that viewpoint, my data really only applies to "picking an ETF with highest 12 month performance momentum", and before that "sector momentum".  There's evidence that others might be doing this better -  the largest momentum ETF ("MTUM" has $8.5 billion in assets) leads the S&P 500 by +3% in the past 3 years:

iShares Edge MSCI USA Momentum Factor ETF ("MTUM"), 3 year return: +16.86% / year
Vanguard S&P 500 ETF ("VOO"), 3 year return: +13.46% / year
5W fund experiment ("5W"?  no real ticker), return since started (3.1 years): +4.32% / year

Maybe it's useful to delve into differences between my experiment and MTUM.  After reading about momentum in academic papers and books, I looked at the description of how MTUM calculates momentum.  They calculate on another level from me - they track performance divided by volatility.  So a stock that has stable high performance gets favored.  They calculate momentum for individual stocks over two time periods (12 mo and ... 6 mo?).  And since they use an outside index, I get more visibility into what they're doing (sadly, Vanguard's momentum ETF provides no visibility - an internal index with no specifics that I've found).

The disadvantages of the 5W experiment are numerous.  I had to buy entire ETFs at a time, with hundreds of stocks inside.  The mixture of stocks has some momentum, but it's diffused by having higher and lower momentum stocks in the same fund.  But it's $0/trade, unlike buying a number of stocks.  Further, my data is 100% from public data on the web - I use etfdb data now, and used to use Google Sheet data on Vanguard's ETFs.  Fairly crude, but I can only sustain the experiment by ensuring it's automated - otherwise the time drain is too much.  I don't factor volatility in at all, so a "lucky" high volatility ETF can get picked over an ETF with a more significant out-performance relative to it's volatility.

My advice to anyone who can't help trying to beat the market as an individual investor:
1) benchmark yourself against the S&P 500.  If you keep losing to the S&P 500, it might be a better investment.
2) read up on the arguments for investing in the S&P 500 (or total U.S. market)... Warren Buffet, A Random Walk Down Wall Street ... any book that uses a lot of stock market history, essentially.
3) make it worth your time - automate your investment approach.  If you're spending hours each week, that's hundreds of hours per year.  Maybe you need to invest less frequently, and automate the selection of stocks.  Otherwise even if you beat the S&P 500 by a tiny amount, you're still out all that time spent.

[MustacheAndaHalf]

The fund is 100% in PSJ (go software, go!).  Next update on May 5th.

[MustacheAndaHalf]

I suspect I need to go back to basics.  The fund changes investments at nearly every update.  Either my timing is perfect (ha!), or the fund is switching investments too often.  Performance so far suggests the latter.  I'll stick to the same approach for this update, but I think "sector momentum" might have been a better idea all along.

So for this update, I looked at the top 12 performing ETFs for the past 12 months, and examined every one of them on morningstar.  Here are the top 5 performers, ranked by momentum:

1) ARKG, 33.45 - (-2.34) = 35.79 mtm   (oops, 88% U.S.)
2) PSJ, 34.81 - 3.77 = 31.04 mtm  (99% U.S.)
3) MJ, 27.27 - (-1.47) = 28.74
4) PXMG, 29.15 - 1.88 = 27.27 mtm
5) XSW, 26.22 - 2.23 = 23.99 mtm

Since ARKG isn't 95% in U.S. stocks, I'm skipping over it and leaving the fund invested 100% in PSJ.  I setup the "95% U.S. equities" rule awhile back to avoid two things: ETFs that tracked commodities, and ETFs that weren't a good fit for an S&P 500 benchmark.  So the fund stays in PSJ.

[MustacheAndaHalf]

I'm bringing the 5W momentum experiment to an end after 3 years and 4 months.  I'm not sure what change is needed, and the fund has already had two different approaches (sector momentum, and then any ETF with momentum).

Since the start of this experiment (Feb 17, 2016) the fund has gained just +17%.  That's 5% annualized, which fell well below it's benchmark of the S&P 500.  The S&P 500 gained +51% over 3.3 years, or about 15%/year.  If there's one takeaway, it's that the S&P 500 beats sector momentum.  Maybe it's also worth noting the 5W experiment beat the S&P 500 early on - a warning that results measured in months might not predict performance over several years.

This experiment doesn't represent all styles of momentum investing.  If anything, it's unique in purchasing entire ETFs at a time, rather than individual stocks.  While the fund did poorly, professionally run momentum funds have done fine - even beaten the S&P 500.  About a year after I started the experiment, Vanguard even created a new momentum fund of it's own.

So the 5W momentum experiment is now closed / over, with a +17% overall performance over 3.3 years compared to the S&P 500 performance of +51%.

[Sciurus]

Thank you for running the experiment and taking the time to share it with the rest of us! 

I have a suspicion that if there is an edge in momentum trading it is already being exploited by algorithms and there may not be room for individual investors to do much.

[arebelspy]

I really appreciate your continual updating and final results on this, despite the outcome. Thank you.

[MustacheAndaHalf]

Glad it was appreciated.  Stick to the S&P 500!

I doubt anyone is trying this exact approach - it's too scattered and unprofessional.  It started by only making trades which cost $0 at Vanguard.  There's no way professional traders or portfolio managers face that restriction.  I also focused on sector momentum, where I bought a Vanguard sector ETF that holds every stock in a specific sector (REIT, utilities, tech, etc).  A professional fund wouldn't need to buy every stock - they could pick the best sector, and buy the highest momentum stocks individually.  Given the restrictions I imposed to keep my costs low, I doubt my approach is competing with professional funds.  Plus, if they followed my approach, they'd probably have closed already with such bad performance relative to both the S&P 500 and other momentum funds.

[Kierun]

Yes, thank you, I enjoyed following along and learning from your experiences.