No. of Recommendations: 141
This is a long post, so here is an executive summary:
- A new method of implementing the seasonal timing idea is presented, which
recommends selling in June and going away, rather than the usual May.
- For the backtest, the total returns are 6.36%/year better than buy and hold.
- On a particular risk metric, this has less than 44% of the risk of buy-and-hold.
- This is specifically recommended for use with the S&P Equal Weight index, tracked by RSP.
- Don't give up, there's a link to a pretty picture further down the message.

I am a big fan of Sy Harding's seasonal timing system.
The general idea can be summarized as follows:
- The market tends to do worse from May to October than the rest of the year.
Nobody knows why, but it was first noticed in the 18th century, and
remains true on average. It's not always true, but often enough to
shift the risk-reward calculation for equities quite a bit seasonally.
- The best date to sell around May is when the market stops rising.
- The best date to buy around October is when the market starts rising again.
Mr Harding's excellent write-up on his system is here:

Mr Harding uses a simple MACD (a technical analysis trend-following
metric) to form the definition of "rising". It works well.

However, I have long wondered if there might not be a better simple metric.
As Zeelotes has pointed out, the number of new highs and new lows on the
Nasdaq exchange lately is an extremely potent stock market predictor.
This can be used as a slower-cycle predictor by looking at which number
is greater than the other, which is the usual method. I call this
the "new highs minus new lows positive" signal. Or, it can be used as a
short-cycle predictor by looking at whether the ratio of new highs
lately to the number of new lows lately is rising. I call this version
the "new highs minus new lows rising" signal. It reacts much more
quickly, and gives a lot more signals per year---it isn't very much
use just by itself for this reason. This signal is implemented
simply by looking at two exponential moving averages (EMA's) and
if the shorter-term one is above the longer-term one, the line
is considered to be rising.

So, here's what I have done-
- I started with Sy Harding's signal generation method.
e.g., around April-May, sell out of the market on the first day
on or after a certain cut-off date that the MACD goes bearish.
- I looked for a short term new-highs-minus-new-lows signal that
has about the same number of signals per year as a MACD.
- I substituted the NH-NL-rising for the MACD. So, we have:
around April-May, sell out of the market on the first day
on or after a certain cut-off date that the NHH-NL-rising goes bearish.
And similarly, around October-November, buy back into the market
when the NH-NL-rising signals goes bullish.

So, this leaves us with a few things to tune.
What is the earliest date in Apr-May that the sell signal can be triggered?
What is the earliest date in Oct-Nov that the buy signal can be triggered?
Which particular moving averages will we use for the NH-NL-rising signal?
What are we trying to maximize?

Let's start with the last one. Since we're looking at a fairly
passive system, with only two trades per year, it seems that a
metric which is fairly "patient" would be good. I used the rolling-year
downside deviation with minimum acceptable return (MAR) of 10%.
This is a metric which says that any 12-month period with a return
over the MAR is risk zero, that any shortfall below 10% has a risk
equal to the square of the size of the shortfall. Thus, a return
of 8% (a shortfall of 2%) is four times as risky as a return of 9%
(a shortfall of only 1%)---double the shortfall, four times the risk.

The second metric is easier: a high long run return for US equities,
measured as the compound annual growth rate (CAGR). But trading what?
The S&P 500 index is dominated by the price movements of a relatively
small number of really huge companies, since the weight of each
company in the index is proportional to the total market value of
the company. I prefer to use the S&P 500 equal Weight index, also
published by Standard & Poors, which weights all 500 of the
companies equally: each one is 0.2% of the index. This index
performs just a tiny bit better on average through the years, and
is not quite as subject to wild gyrations. It's also a little more predictable.
So, the returns quoted here are for the S&P 500 Equal Weight.
I have included dividends. This index is easy to track by buying
the Rydex exchange traded fund with ticker RSP. For pricing data,
I have used the actual prices of RSP for its history since 2003.
Prior to that, I used the "official" S&P equal weight total return
index from S&P back to 1990. Prior to that, I have used another
reputable source for a reconstructed version of the same thing.
This earliest data assumes that you have exactly equal weights
every day, which isn't true of the official index which rebalances
quarterly. Therefore the returns in the earliest years of this
test are slightly higher than what you would have had in real life.

So, now all of that groundwork is out of the way, here are the
results of the test.

Buy and hold the S&P 500 Equal Weight Total Return index
Test period Jan 3 1978 to June 2 2009
CAGR 13.24% Risk based on downside deviation = 11.037%
Probability of a positive rolling year 80.37%
Worst rolling year -51.67%
Ulcer Index 11.70%

Seasonal system (two trades per year):
Long winter S&P 500 Equal Weight Total Return, cash summer
Test period Jan 3 1978 to June 2 2009
CAGR 18.43% Risk based on downside deviation = 4.733%
(42.88% of the risk of buy-and-hold, 6.47%/year higher returns)
Probability of a positive rolling year 90.86%
Worst rolling year -24.88%
Ulcer Index 6.43% (54.9% of the risk of buy-and-hold on this risk metric)
Percent of the time long the market: 64.4%

Here are all the magic tuning values used in this test:
- Earliest sell date at market close June 5th
- Earliest buy date at close October 14th
- Data source for the new highs and new lows: Pinnacle Data file B2 (Nasdaq)
- For each day, calculate the % of Nasdaq issues which hit new 52-week
highs on that day minus the % of issues that hit new lows.
Calculate two exponential moving averages of this series of values.
- The two EMA's used are 5 and 17 days. 5&18 is about equivalent.
- If the EMA5 is above the EMA17, it is bullish. I also counted
it bullish even if it's a really tiny bit lower, by up to 0.05%.
This is probably just random noise, but I mention it for completeness' sake.

Here is a graph of the portfolio values for buy and hold (blue) as well
as for the cash-in-summer approach described here (pink).
In winter the two are the same, holding the S&P equal weight, and in
summer the trading system holds cash (3 month T-bill rates).

I highlighted a few of the years that the system really made things worse.
However, note that the returns were worse, but still positive/good in these particular years (2nd column).
That's no guarantee for the future---nothing works all the time.
The strength of the system seems to be in its ability to dodge a really
large summer loss from time to sime, such as 1978, 1981, 1987, 1990, 1998, 1999, 2001, 2002, and 2008.
(and apparently also 1971 and 1974, though they aren't in this test)
Conversely when it doesn't work and you're worse off, the year is
usually still good, and it's pretty rare to be worse off by very much.

Year   Buy and hold   Seasonal portfolio    Advantage   Sell at close on   Buy at close on
1978 11.6% 22.0% 10.5% 1978-06-05 1978-11-09
1979 26.2% 28.5% 2.3% 1979-06-19 1979-10-31
1980 34.8% 23.3% -11.5% 1980-07-28 1980-10-14
1981 5.0% 22.7% 17.7% 1981-06-05 1981-10-14
1982 29.2% 3.5% -25.7% 1982-06-07 1982-10-14
1983 33.0% 34.8% 1.8% 1983-06-08 1983-11-10
1984 3.5% -3.7% -7.2% 1984-07-06 1984-10-15
1985 32.3% 37.2% 4.9% 1985-06-07 1985-10-14
1986 22.1% 26.3% 4.2% 1986-06-09 1986-10-14
1987 10.3% 35.6% 25.3% 1987-06-30 1987-11-02
1988 17.0% 16.1% -0.9% 1988-06-28 1988-10-18
1989 30.6% 33.5% 2.8% 1989-06-09 1989-11-10
1990 -14.2% 14.6% 28.8% 1990-06-07 1990-10-18
1991 36.4% 36.8% 0.4% 1991-06-07 1991-10-15
1992 15.6% 17.6% 2.0% 1992-06-09 1992-10-14
1993 15.0% 9.8% -5.2% 1993-06-08 1993-10-14
1994 1.3% -1.1% -2.4% 1994-06-20 1994-10-14
1995 33.0% 26.8% -6.1% 1995-06-27 1995-10-16
1996 17.2% 17.7% 0.5% 1996-06-05 1996-10-15
1997 30.3% 24.9% -5.4% 1997-07-21 1997-11-21
1998 12.0% 35.3% 23.2% 1998-06-05 1998-10-14
1999 9.5% 22.2% 12.7% 1999-06-14 1999-10-26
2000 8.9% 10.8% 1.9% 2000-07-21 2000-10-20
2001 2.7% 21.0% 18.4% 2001-06-08 2001-10-15
2002 -15.5% 13.7% 29.2% 2002-06-05 2002-10-14
2003 35.3% 24.6% -10.7% 2003-06-11 2003-10-14
2004 15.4% 12.8% -2.5% 2004-06-14 2004-10-27
2005 10.8% 9.8% -0.9% 2005-06-24 2005-10-24
2006 13.1% 6.7% -6.4% 2006-06-06 2006-10-16
2007 -0.6% 3.2% 3.8% 2007-06-06 2007-10-31
2008 -37.0% -2.1% 34.9% 2008-06-09 2008-10-15

Some observations:
In my testing this approach works very much better with the S&P
Equal Weight index than it does with the standard S&P 500 index.
Both work, but the equal-weight works much better.
Trading the S&P total return index, performance increases by about "only " 1%/yr.
So, if you're going to use this system with one hold, buy RSP in winter not SPY.
However, it should add value for any portfolio concentrated in US large caps.

There is a quick and dirty way to find out the current NH-NL-rising signal.
Go to this URL each day starting on the earliest signal date.$NAHL&p=D&yr=0&...
If the blue line is higher than the red line, the signal is bullish.
(If the two are extremely close to tied, call it bullish)
You won't get exactly the same result, since different data sources have
slightly different counts for Nasdaq highs and lows, but the end result
will probably be so similar that any difference will be statistical noise.

Note that this variant of the system is in the market a fair bit
longer on average in May-June than Sy Harding's approach. This is
particularly useful for those who are using mechanical investing (quant)
approaches, as these tend to do quite well on average in May.
It's also good for anyone who just plain likes to be in the
market a larger fraction of the year.

What has it done for me lately? This was a harsh winter, right?
The year ending June 1 2009 returned +11.9%, versus -28.8% for buy-and-hold on the same index.
So, even though there was a big dip during the winter, it was regained.

All figures in this write-up include dividends.
However, I didn't include trading costs. These days, trading RSP at
Interactive Brokers costs about 0.052% per trade including the bid/ask
gap and the commissions, which adds up to about 0.1%/year, definitely under 0.2%.
No big deal.

If used over a long period of time, I very much believe that this
seasonal approach will reduce risks by a large amount, at very little or
no long run average cost. In testing the long run cost was negative,
i.e. it actually improved returns a lot (over 6%/year). However, even
without that it's very much worthwhile given the huge risk reductions.


PS, this is my 5000th post!
Print the post  


When Life Gives You Lemons
We all have had hardships and made poor decisions. The important thing is how we respond and grow. Read the story of a Fool who started from nothing, and looks to gain everything.
Contact Us
Contact Customer Service and other Fool departments here.
Work for Fools?
Winner of the Washingtonian great places to work, and Glassdoor #1 Company to Work For 2015! Have access to all of TMF's online and email products for FREE, and be paid for your contributions to TMF! Click the link and start your Fool career.