A few sell in may articles…

Have you read any “sell in May and Go Away” articles? These articles suggest that returns from May to October are terrible. So bad, in fact, that you’d be better off just avoiding the whole mess altogether. I’ve included some of the better-written posts for your reading pleasure.

Forbes: Sell in May and Go away?

Adam Sarhan opines that a bad jobs report and lackluster earnings will cause problems. But his title is clearly referring to the seasonal pattern.

Wikipedia: Sell_in_May

It covers research in many countries and is enough to convince me to dig further.

TheStreet: the-real-reason-we-say-sell-in-may-and-go-away

Ken Roberts describes a difference in sectors, with retail, media, utilities, and defensive stocks enjoying the summer more than others.

CFA Institute: Sell in May and Go Away?

Joachim Klement, CFA, argues that it works. Of course, he does has an ulterior motive, he wants a summer vacation

All this reading made me very curious, and so I wanted to know…

Does sell in May actually work?

The baseline

First, I create a screener that rebalances on the first of May, and the first of November. It uses the typical universe, where all stocks below $500,000 trading volume are excluded. We do this as a baseline, so we don’t get tricked by rebalancing effects.

Figure Rebalance May 1st and November 1st

Seems pretty much like a baseline to me. Let us knock out May through October. Notice the flat lines that run from May until November.

Figure Sell in May and go away

There’s something deeply pedantic in my desire to benchmark the other side of this coin. Nevertheless, here it is.

Figure Buy in May and stick around

Lots of time in the market gives you very little reward…

Is May itself special?

What if you just held stocks for June through October. Will you still have poor performance? This might happen to you if you read all the Sell in May articles people publish, cogitate, and realize much of May has left.

Figure Buy in June and stick around

Yup — still works. That’s one resilient strategy. It works even better than the original! It turns out, though, that you should buy in November. Waiting until December doesn’t work as well.

May in various sectors

Does this effect vary by sectors? Does it work in reverse for some sectors? If so, that would be handy, since we’d buy summer stocks in summer, and winter stocks otherwise.

First up, utilities (the first one is the summer performance, the other one is the winter performance):

Summer utilities

Winter utilities

So with an average utility stocks, you are better off holding the stock during summer than selling it. Still, most of the gains occur during the winter months.

Other sectors

Hmm… So if you should hold on to your utility stocks, does that mean there are sectors where the sell-in-may effect is even worse? The answer is yes! Otherwise the Sell in May effect would be known as the Hold and Underperform Mildly in May effect. First up, we highlight tech stocks (summer first, winter second).

Summer technology

Winter technology

With technology stocks, holding all such stocks in the summer months would have lost you 50% of your money! Though, there are sectors that are even worse during the summer…

Summer transportation

Winter transportation

When holding transportation stocks only in the summer, you lost a 64% of your initial value. Eeeks! In fact, when we ran our simulation, it was a bad idea to hold stocks in the following sectors: Technology, Transportation, Basic Materials, Energy, Industrials, or Consumer Cyclicals sectors. All of these sectors distinguished themselves by actually losing money during the summer. It almost seems like there’s a way to measure the seasonality of an industry or group. If so, we can sort our portfolios to avoid the seasonal industries, market cap segments, etc when the season is wrong.

Update one…

It’s still May, so I thought I’d write more about the Sell in May and Go Away effect. ​​ Why am I doing this? ​​ I’m writing about sell in May because it’s May (we learned above that Sell in June works just as well), and I want to know if I have to change my strategy. ​​ In years past, I’ve pooh-poohed the notion as data mining: a hack that fits the data, but would never work going forward. ​​ As you can tell from the last article, I’m figuring out whether I was wrong, and there really is a seasonal bias to returns. ​​ If there is, I want to know that so I can exploit it. ​​ I have a few other questions.

  • ​​ Is it a vacation effect? ​​ Perhaps all the people going on vacation kills the market. ​​ 
  • Do we have the right collection of months? ​​ If we do, it will be easier for us to come up with an explanation for them
  • Is really valid? ​​ It seems too simple to be true.

The Vacation hypothesis

To test the vacation hypothesis, we need a slightly different rig. ​​ Initial ideas involve rebalancing monthly, then weekly, etc, and filtering out different time periods. ​​ But all that turns out to be complexity for no reason. ​​ The following panel shows us what we want to do – simply define a set of time periods that we want to exclude, and we rebalance anytime we enter or leave any such period. ​​ This could be easily extended to market conditions, phases of the moon, etc.

So, before we started this tweaking, we had the following panel:

This got us the following results. ​​ 

When we included all sorts of holiday periods (the periods right before the NYSE holidays, excluding Good Friday because it seemed to move all over the calendar), we got

which in turn gave us these results.

Neither one loses every year, but with holidays seems less bad than without holidays (the little bars are each year vs the S&P 500, which is brown). ​​ Next we start taking holidays out of our vacation collection, starting with Christmas, Thanksgiving, MLK day (in January), Washington’s Birthday(in February) , ​​ Memorial day, Labor Day, and finally July 4th. ​​ When we exclude a vacation that falls in the summer months, we assume we buy stocks for that period, and then sell them again. ​​ Time for a table:

Excluded from vacation (includes all above)Monthly Average Return
Nothing.42%
Christmas0.2% (half the performance!)
Thanksgiving0.04% (down by 75%!)
January-0.04% (now we’re losing money!)
September-0.08% (double down)
May-0.08% (makes no difference)
July​​ -0.07% (slight bump – in the wrong direction!)
February0.02%  ​​​​ (wait – what!)
INCLUDE second half of Feb0% (there goes the January effect ends sometime in Feb hypothesis)
“Optimal” vacations (exclude all but February vacation, July, summer is from May to November)-0.08% (repeated from the “September” column)
good vacations (exclude all but February vacation, July, summer is from June to November)-0.13% (Sell in June, May is too soon!)
“Optimal”: vacations (exclude all but February vacation, July, summer is from June to October)-0.31% (Don’t wait until Halloween to buy – or you’ll miss the treat!)
“Simplified”: Use four months as vacation (heck with these holidays)-0.18% (It’s more important to get the months right, than it is to get the holidays right)

As you can see, performance goes down as we progressively exclude more vacations – until May. ​​ This kills the idea that vacations, per se, are bad for the market, and leads us with the odd idea that winter vacations are buying opportunities, while summer vacations are not. ​​ 

Is it valid?

The oddly specific vacations make me worry. ​​ Considering February, I start to worry about overfitting. ​​ What’s going on in February to cause declaring that one week (from the 14th to the 20th) a vacation to make huge difference. ​​ To help determine this, I use the backtest tilt twice – once for the vacation against all stocks, and once for the vacation against the “simplified” summer screen.

To summarize the chart, I look at two key features: the lack of huge outliers, and the presence of a large number of dots scattered on both sides of the diagonal line. ​​ These both conspire to tell me that the vacation underperformance was not likely due to a fluke, but rather some mysterious, persistent effect. ​​ If, for instance, you look at the X axis, we see all the times our summer strategy sat out vs the market. ​​ We see a red dot at nearly 20, but another green dot at almost -20. ​​ The real story lies in the numbers: the blue line traces out how many dots are above or below the diagonal line, and we lose more by stepping out than we gain by not getting hurt. ​​ This turns out to be exactly what we want, in this case.

These two charts show you the impact of losing half your money in 20 years while the market returned a bit more than 500%. ​​ You can see that the last two years have borne out the theory, though the two years before that were not good years to step out in the summer.

A FAQ Interlude

Here I’m going to deal with questions I had while trying to convince myself this works.

Does the fact we have more trading days in the non-vacation throw our analysis off?

If we were dealing with a lower positive performance, this would be a problem. ​​ Sugata Ray shows how to deal with this problem when he analyzes the January Effect. ​​ In our case, though, the vacation time period actually manages to lose money. ​​ The fewer days it can lose money in, the more powerful the effect.

You have all sorts of stocks in the screen! ​​ What if this effect only works on micro-caps?

We’ll attack that in more detail later on, but the short answer is no: unless you invest in mega cap stocks (more than 50B) you will lose money. ​​ And even investing in mega caps only got you back to even. ​​ We attack this problem later when we answer where can you hide in summer to avoid this effect. ​​ So, stay tuned!

Is performance is because of a few really bad summers?

​​ If so, that would be bad. ​​ We analyzed overfitting above for the holidays, but if the whole strategy is overfitted, then we are in trouble . ​​ To analyze this, we could look again at our backtest by time chart (above), and see that there were four years where the S&P 500 managed to turn in a negative year (five for our all stocks screen), while the summer strategy turned in a negative year 11 times out of the last 20. ​​ 

What about the wacky rebalancing?

Is my (admittedly odd) way of rebalancing the source of all this underperformance? ​​ Let’s find out, by rebalancing monthly (still staying out of the market for the “vacation” months). ​​ This has the side effect of killing the holiday weeks, as they don’t start at the beginning of the month. ​​ But it doesn’t matter: the result is still negative.

What if we’ve already had a bad month in X?

What happens if May was a terrible month before summer starts? ​​ Or if January wasn’t very nice? ​​ Or if the Santa Claus rally was canceled? ​​ Or Halloween was more of a trick than a treat? ​​ No worries! ​​ We test whether this screen holds if the month in question turned in below average, above average. ​​ The average in question is the 20 year average of the SPY’s return in the month in question. ​​ 

MonthSPY averageAboveBelow
January-0.04-0.19-0.21
February-0.04-0.16-0.12
March1.57-0.25-0.15
April1.99-0.26-0.14
May0.40.12-0.51
June-0.25-0.01-0.24
July.29-0.16-0.24
August-1.32-0.18-0.22
September-0.30-0.21-0.07
October1.80-0.41-0.02
November1.69-0.22-0.18%
December1.26-0.18-0.22
    

This is pretty spooky – although the good may is still half the performance we might expect if there were no effect at all. ​​ As an effort to codify this, and to detect overfitting, I created the seasonality score: ​​ basically how subject to the vacation effect the market should be given the performance of the past May, June, October and September. ​​ The seasonality score hands out points as follows (we can’t use the above/below average in our score since that would be using future knowledge):

  • One point each if may and june are not above 1%
  • One point each if may and June are below -1%
  • One point each if September and October are above 1%
  • One point each if September and October are above -1%

Basically it boils down to: ​​ if performance is good in May and in June, or performance is bad in September and October, we should be skeptical of seasonality.

Maybe it’s just the last 20 years?

Perhaps, but looking back three centuries, it seems to persist. The only period the summer didn’t have negative returns was for 1851-1900. ​​ I suppose we won’t ever know for sure, but that’s good enough for me.

Resigned to the summer blues

Now that we’ve decided that the effect is real, what do we do? ​​ My first instinct is to hide, that is, to find stocks for which this effect does not hold true. ​​ To make the standard clear, I create a grading system (using the period 1996-present):

  • Anything that outperforms the S&P 500 in the summer gets an A. ​​ This is slightly trickier than it seems: ​​ we don’t insist that it do better in 87 trading days than the S&P 500 does in 252. ​​ So, what to do? ​​ Just divide the S&P 500’s monthly performance (0.64%) by 252/87. ​​ Now 252/87 is darn close to 1/3, so we say that anything that performs better than 0.21% monthly in the summer months gets an A. ​​ A solid, shining A. ​​ As you can tell, this is not an easy bar to meet.
  • Anything that gets about half that performance is still a reasonable investment during the summer, even if it’s not as nice as the winter. ​​ So, anything that gets 0.1% monthly performance gets the B it deserves. ​​ 
  • Anything that gives a positive return, no matter how small, can be an investment, even if it’s not a very good one. ​​ That sounds like a C to me.
  • It might make sense to hold during the summer even if you lose up to 0.1% monthly by doing so. ​​ You’d only lose 21% over a 20 year holding period. ​​ There might be lots of reasons to hold your nose and your overcooked stocks if this was all you had to worry about. ​​ Taxes, for instance, are more than enough reason to hold on through the summer.
  • We’ve got a benchmark of -0.38% monthly (from 1996), as the result of holding everything (from 1995-present). ​​ If the screen beats that, it gets an E – still a bad idea, but not so special it gets the dreaded F.
  • Self explanatory – these strategies are even worse than the benchmark. ​​ These may be good candidates for shorting.

Why did we do this? ​​ So we can quickly compare a bunch of hiding places.

Eat your brass tacks first (or don’t!)

​​ Now, if you were paying attention, you may have noticed that I used -0.38% as our benchmark, rather than the -0.31 earlier. ​​ We also reduced the S&P 500 line from 0.74 monthly to 0.64%. ​​ I’m sure this gives you a gentle desire to know what’s going on. ​​ Wonder no longer – we’ve excluded 1995, which was a good year. ​​ This knocks down both the S&P 500 and vacation performance. ​​ We do this because we try a variety of things that look back, and we don’t want to skew the results by looking back beyond the beginning of our data.

A large hiding place?

Do larger stocks survive better in the heat? ​​ Let’s try a range of market capitalizations (we use the range 1996 to present, from here on out) and see:

Market Cap(63 day median)ChartGrade
Greater than 100bD (-0.1%)
Between 10B and 100BE (-0.24%)
1B to 10BE (-0.33%)
100M to 1BE (-0.3%)

It appears there’s no place to hide here! ​​ As a Hail Mary, we try the largest 10 companies by market cap, only to find out it does no better than the mega cap row.

Does beta lead to summer alpha?

Maybe the market punishes itself in the summertime, and by avoiding stocks correlated with the market we can dodge the bullet. ​​ Let’s see (same 1996-present timeframe, all caps included, charts omitted since they all look very similar).

Beta Rank(rank of 63 day median,Beta > 0)Grade
Between 20 and 80F (-0.42%)
Less than 10E (-0.31%)
Between 10 and 20D (-0.07%)
Between 20 and 30E (-0.22%)
Between 30 and 40F (-0.43%)
Between 40 and 50F- (-0.6%)
Between 50 and 60F (-0.48%)
Between 60 and 70F (-0.43%)
Between 70 and 80E (-0.34%)
Between 80 and 90F- (-0.61%)
Greater than 90E (-0.34%)
NegativeE (-0.26%)
Default Vacation(no beta monkeying)E (-0.31%)

I’d say not – the performance looks like an W, with the troughs being in the 40-50 range and in the 80-90 range. ​​ I forsee an article about beta, and only beta, but for now, I can see it as a component of a summer short screen, or as a component of a hiding place. ​​ But it’s not enough by itself.

Can we leverage leverage?

Perhaps debt makes companies vulnerable to bankers, who, bothered by the summer heat, call in more loans and enforce more debt covenants more strictly. ​​ If so, I’d expect to see a spike of negativity among highly indebted companies, and less gloom elsewhere.

Leverage Rank(rank of total liabilities 1Q / total assets 1q)Grade
Between 20 and 80E (-0.35%)
Less than 10F- (-0.62%)
Between 10 and 20F (-0.49%)
Between 20 and 30F (-0.43%)
Between 30 and 40E (-0.38%)
Between 40 and 50F- (-0.42%)
Between 50 and 60F (-0.32%)
Between 60 and 70E (-0.28%)
Between 70 and 80E (-0.3%)
Between 80 and 90E (-0.2%)
Greater than 90F (-0.38%)
Default Vacation(no leverage monkeying)E (-0.38%)

Apparently not. ​​ Still no place to hide, but at least we know that more leverage is better than less. ​​ Counterintuitive as it seems, having liabilities is good during the summer months. ​​ Perhaps companies that are stable enough to loan money to are stable enough to survive the tough times, seasonal though they may be. ​​ Or perhaps the large enterprise value relative to the market capitalization acts as an anchor.

How about quarterly standard deviation?

OK, let’s look at what actually happened in the last quarter. ​​ Perhaps we can do better during rough times by finding stocks that are buffeted around less. ​​ We rank the standard deviation of the 1 day change in price, taken over the last 63 trading days (one quarter). ​​ 

Stddev Rankrank of (stddev (change of close over 1 day) within 63 days)Grade
Between 20 and 80E (-0.34%)
Less than 10C (0.02%)
Between 10 and 20D (-0.07%)
Between 20 and 30E (-0.19%)
Between 30 and 40E (-0.26%)
Between 40 and 50E (-0.3%)
Between 50 and 60E (-0.33%)
Between 60 and 70F (-0.44%)
Between 70 and 80F (-0.58%)
Between 80 and 90F- (-0.72%)
Greater than 90F- (-0.97%)
Default VacationE (-0.38%)

Now we are cooking with gas! ​​ This shows a beautiful dosage response curve, and the last two entries look like good short screen candidates.

Can momentum keep us moving?

Perhaps this is a Momo effect, if not a Ducky Momo one (Sorry about the Phineas and Ferb callout…)? ​​ To test this we check quarterly price performance, ranked in deciles. ​​ Classically when doing momentum studies, academics will exclude the last month from consideration. ​​ We’ll call that “momentum”, as that’s a nice fancy word. ​​ The price performance including the last month will use the “momo” moniker.

Performance rank(Momo or momentum)Momo Grade(including last month)Momentum Grade(last month excluded)
Between 20 and 80E (-0.29%)E (-0.25%)
Less than 10F- (-1.12%)F- (-1.04%)
Between 10 and 20F- (-0.66%)F (-0.58%)
Between 20 and 30F (-0.53%)E (-0.37%)
Between 30 and 40E (-0.37%)E (-0.34%)
Between 40 and 50E (-0.28%)E (-0.27%)
Between 50 and 60E (-0.25%)E (-0.25%)
Between 60 and 70E (-0.18%)E (-0.24%)
Between 70 and 80E (-0.17%)E (-0.27%)
Between 80 and 90E (-0.16%)E (-0.25%)
Greater than 90E (-0.23%)E (-0.32%)
Default VacationE (-0.38%)E (-0.38%)

More gas! ​​ We learned two things: ​​ momentum works, and it is better not to exclude the last month. ​​ 

How about industry momentum?

This one’s slightly tricky: how do we wrap our heads around it? ​​ We take each stock, and mark it with the performance of it’s industry over the last year. ​​ Having done that, I then rank these and sort them into deciles, just as we did with the others. ​​ We have two different industry momentums: one for yearly industry performance, and one for quarterly industry performance.

Industry Performance rankYearlyQuarterly
Less than 10F- (-0.82%)F- (-0.77%)
Between 10 and 20F- (-0.57%)F (-0.72%)
Between 20 and 30F (-0.58%)F (-0.46%)
Between 30 and 40E (-0.37%)E (-0.37%)
Between 40 and 50E (-0.52%)E (-0.55%)
Between 50 and 60E (-0.18%)E (-0.33%)
Between 60 and 70E (-0.21%)E (-0.27%)
Between 70 and 80E (-0.48%)E (-0.22%)
Between 80 and 90E (-0.08%)E (-0.1%)
Greater than 90E (-0.30%)E (-0.24%)
Default VacationE (-0.38%)E (-0.38%)

More gas! ​​ We learned two things: ​​ momentum works, and it is better not to exclude the last month. ​​ 

Can four wise men save us?

The four wise men in question are Joseph Piotroski, Masood Beneish, and Altman, and James Montier, and their scores are (respectively) the Piotroski F score, the Beneish M Score, the Altman Z Score, and the Montier C Score. ​​ Time for a big table:

RankAltman Z-scorePiotroski-1 * Beneish-1 * Montier
0-20E (-0.26%)F (-0.54%)F- (-0.64%)F (-0.52%)
20-40E (-0.35%)F (-0.56%)E (-0.37%)E (-0.19%)
40-60F (-0.41%)E (-0.36%)E (-0.25%)E (-0.36%)
60-80F (-0.39%)E (-0.27%)E (-0.25%)E (-0.35%)
80-100F (-0.43%)E (-0.24%)E (-0.33%)E (-0.26%)

It appears the Z score is worthless – but the others all seem to work more or less. ​​ None of them, however, earn us the coveted C (did I just say that?!?!).

How ‘bout those fancy scores you have under sauce?

The four wise men in question are Joseph Piotroski, Masood Beneish, and Altman, and James Montier, and their scores are (respectively) the Piotroski F score, the Beneish M Score, the Altman Z Score, and the Montier C Score. ​​ Time for a big table:

RankAltman Z-scorePiotroski-1 * Beneish-1 * Montier
0-20E (-0.26%)F (-0.54%)F- (-0.64%)F (-0.52%)
20-40E (-0.35%)F (-0.56%)E (-0.37%)E (-0.19%)
40-60F (-0.41%)E (-0.36%)E (-0.25%)E (-0.36%)
60-80F (-0.39%)E (-0.27%)E (-0.25%)E (-0.35%)
80-100F (-0.43%)E (-0.24%)E (-0.33%)E (-0.26%)

It appears the Z score is worthless – but the others all seem to work more or less. ​​ None of them, however, earn us the coveted C (did I just say that?!?!).

How about fundamental hot sauce?

It turns out we have a set of four scores we have used to create a number of screens. ​​ One analyzes balance sheets, one cash flow statements, one income statements, and the last valuation. ​​ They are all modeled off of Piotroski score, in that they each have a checklist, and you get one point for each item that conforms. ​​ Since we did academic scores, and found them effective, we decided to do the same for our fundamental scores. ​​ Time for another big table:

RankIncome StatementCash FlowValue ScoreBalance sheet
0-20F (-0.55%)F (-0.52%)F (-0.5%)F (-0.58%)
20-40F (-0.54%)F (-0.47%)E (-0.34%)E (-0.42%)
40-60F (-0.43%)E (-0.37%)E (-0.37%)E (-0.25%)
60-80F (-0.25%)E (-0.35%)F (-0.39%)E (-0.34%)
80-100F (-0.43%)E (-0.24%)E (-0.33%)E (-0.26%)

None of these scores are impressive, but they all do seem to do worse when low, and better when high.  ​​​​ 

Technical Scores

It turns out we have a set of four scores we have used to create a number of screens. ​​ One analyzes balance sheets, one cash flow statements, one income statements, and the last valuation. ​​ They are all modeled off of Piotroski score, in that they each have a checklist, and you get one point for each item that conforms. ​​ Since we did academic scores, and found them effective, we decided to do the same for our fundamental scores. ​​ Time for another big table:

RankFear FactorJoy FactorWrong Score
0-20C (0.0%)D (-0.01%)F (-0.1%)
20-40D (-0.05%)D (-0.1%)E (-0.27%)
40-60E (-0.16%)E (-0.16%)E (-0.33%)
60-80F (-0.46%)E (-0.37%)F (-0.53%)
80-100F- (-0.8%)F- (-0.7%)F- (-0.84%)

None of these scores are impressive, but they all do seem to do worse when low, and better when high.  ​​​​ 

Does the sector defence work?

I suspect you all were waiting for me to get around to this one. ​​ Do different sectors respond differently to our vacation times? ​​ Yes they do. ​​ We can test this by backtesting everything during vacation times, and heading over to the positions by category tab:

Taking this shortcut lets us see that nothing (other than the few oddball Unknown positions) will backtest with a positive return. ​​ Why should this work? ​​ We are scoring the average return of each position as we sweep through the backtest. ​​ In theory, that should line up with reality. ​​ Let’s verify that:

SectorGrade
TechnologyF (-0.63%)
EnergyF (-0.59%)
Financial ServicesE (-0.15%)
IndustrialsF (-0.56%)
Basic MaterialsF (-0.67%)
Communication ServicesE (-0.46%)
HealthcareE+ (-0.11%)
UtilitiesC (+.01%)
Consumer CyclicalsE (-0.34%)
Consumer DefensiveE (-0.14%)
Real EstateE (-0.15%)
REIT’sD (-0.08%)
Limited PartnershipsC (+0.02%)
ADR’s (American Depositary receipts)F (-0.4%)

As it turns out, they don’t exactly match up, but they are very close. ​​ The differences lie in that when we did the positions breakdown, we implicitly gave the different timespans different weights, based on how many companies were in that industry. ​​ So, for instance, if there were relatively few industrials during the 90’s, and they did poorly then, that might explain why they would do worse when backtested by themselves.

How about testing all 134 industries?

No. ​​ Just no. ​​ That would be one looooong table. ​​ Also, some of those industries have very few companies, and would be all about those companies, rather than about some feature of the market or economy.

Putting it together – some long strategies

The approach of maximum utility

We’ve been through a lot so far: ​​ walls of tables, inscrutable graphs, and more backtests than you can shake a stick at. ​​ Now, it’s time to put what we’ve learned so far together, and see if we can get a C or better, in a reasonably diversified, sane portfolio. ​​ Looking backwards, scanning for C’s, I find the following:

  • ​​ Utilitiesdsadasdsads
  • Limited Partnerships
  • Fear Factor rank 0-20
  • Stddev rank 0-10

Ok, that’s a game plan. ​​ It seems to do surprisingly well, 0.14% monthly, which is a solid B!

The only problems I have are twofold:

  • ​​ It does not even remotely resemble diversified: ​​ all its picks are utilities! ​​ This is in one sense a minor objection, compared against the alternatives. ​​ But, my luck is such that just after everyone follows this strategy, Utilitigate will happen, and all utilities will crash. ​​ 
  • Performance during the winter is not as stunning as some featured screens. ​​ I think this is very much a “First World Problem”, though: