The SweetSpot Investment Letter

October 25, 2006

Subject:  back-test

Greetings,

Key question:  How would we have fared trading a "SweetSpot-style strategy" (try saying that five times really fast) from the earliest moment it could have been traded?   The answer gives us our “baseline” numbers, which Excel calculated for us by applying the original SweetSpot formula to annual data covering the 17-year period from 1989 through 2005.  Our baseline universe consists of 45 completed three-year trades.

Baseline Returns:

Average return per three-year trade:  54.2%

Average market return during same period:  45.6%

Average excess return per three-year trade:  8.6 percentage points

Looking at the annual performance of completed trades plus current holdings, our universe is 144 “position-years”:

Average return per position-year:  15.5%

Average market return during same period:  12.5%

Average excess return per position-year:  3.0 percentage points

The answer to our key question is that we would have done well trading this strategy.  We beat the market over time, [1] something only Warren Buffett, Bill Miller, and a few select others have been able to do.  Given SweetSpot's real-time performance, a back-test showing any long-term advantage over the market must be considered a favorable outcome.

Years One, Two, & Three

Did our average performance vary among Years One, Two, and Three of a trade? [2] Yes, and in a surprising way:  Year Three has been our best year, up almost 20% on average.  Year One has been least profitable, actually losing to the market by a small margin (but still up double digits).  Luckily, we don’t care about Year-One performance except when it lets us buy more shares cheaper.  And now we know we can add shares with confidence at any time during our three-year holding period.

Strategic Improvements

Of course, there is always room for improvement.  Based on additional back-testing, a few refinements will be implemented beginning with our 2007 trades.  Henceforth we will:

-- Buy the top three funds picked by either of two formulas;

-- Buy any “also-rans”; and

-- Abandon sell signals.

This means we may buy more than three funds each year.  Historically, taking this approach we would have held from nine to 13 funds at any given time.

Formula One versus Formula Two

In 2004 I accidentally used the “wrong” formula (“Formula Two”) to identify our picks.  My mistake had something to do with denominators... Anyway, according to our original formula (“Formula One”), we should have bought telecom and not energy service in 2004.  That was a fortuitous screw-up...

When I looked more closely, I found that Formula One favors funds that performed well even in the face of heavy cash outflows, whereas Formula Two favors funds that performed poorly as expected.  So one formula likes "momentum" and the other likes "value."  Don't be misled, however.  SweetSpot is in no way a momentum strategy.  All SweetSpot sectors are value sectors; some simply show better price performance than others before we buy in.

More than 80 percent of the time the two formulas picked the same funds; they differed only ten times.  And they both beat the market over time by roughly the same margin.

Also-Rans

Seven times from 1989 through 2005 the number-four pick (which wasn’t bought) had cash outflows that were less than one percent behind the number-three pick (which was bought).  Is this a meaningful difference?  The numbers said no:  As a group, the seven “also-rans” were profitable and beat the market by even more than the top-three picks.

Additions to Baseline Numbers

We need a sample size of at least 30 for our results to give us confidence that the numbers reflect our true prospects.  We get there by combining Formula Two picks (when different from Formula One) and also-rans.  Both are in the category of “near misses” in that they were among the leaders identified by our original formula, just not in the top three.  When I broke down these trades into position-years, we had a sample size of 38:

Average return per position-year:  18.0%

Average market return during same period:  11.5%

Average excess return per position-year:  6.5 percentage points

Inexplicably, these funds more than doubled the margin by which our baseline trades beat the market.  Apparently a near miss is better than a hit.  In any event, we know that buying additional funds would increase our level of diversification, thus lowering our investment risk.  Given these numbers, we would achieve this lower risk while at the same time enhancing our expected returns.

Consolidated Returns

When we add also-rans and Formula-Two picks to our baseline trades, we get our “consolidated” results.  These are more indicative of our prospects than the baseline numbers; they reflect our strategy as we will be trading it henceforth.  Of 57 completed three-year trades:

Average return per three-year trade:  54.7%

Average market return during same period:  44.4%

Average excess return per three-year trade:  10.3 percentage points

Annualized, we have 182 position-years:

Average return per position-year:  16.0%

Average market return during same period:  12.3%

Average excess return per position-year:  3.7 percentage points

Many happy returns, but we may be able to do even better.  Note that holding nine or more funds at any given time is a key feature of our strategy.  The first year we could have held nine or more funds was 1991.  If we test performance beginning in 1991, we have 50 completed trades:

Average return per three-year trade:  56.3%

Average market return during same period:  43.8%

Average excess return per three-year trade:  12.5 percentage points

Annualized, we have 172 position-years:

Average return per position-year:  17.4%

Average market return during same period:  12.6%

Average excess return per position-year:  4.8 percentage points

Spreadsheets are fun...

Sell Signals

From 1999 until 2005, we sold a position one or two years early if it became a cash-inflow leader one or two years after we bought it.  In 2005 I discovered that despite a strong beginning, our sell signals did not improve our results over time.  Not that they hurt us – they just didn’t help.  Our baseline back-test did not incorporate a sell signal; rather, we simply held each position for three years.

But how well did sell signals perform in the long run?  Again, our sample size was small.  If we count all trades from 1989-2005, whether based on December 31 or November 30 numbers [1], there have been 13 sell signals, or 19 position-years.  Even with such a small sample size, it’s hard to ignore the consistency of the numbers:

Average return per position-year:  10.6%

Average market return during same period:  6.9%

Average excess return per position-year:  3.7 percentage points

Funds we would have held but for a sell signal beat the market by the same margin as our consolidated trades and within a point of our baseline number.  The lesson:  As a group, positions that generate “sell signals” are no different from those that don’t.  Sell signals have no value, so they can’t be worth the hassle they entail. [3]

As always, I welcome comments and questions.

Cheers,

Neil

[1] As good as these numbers are, they don’t even include our first three years of real-time trades (1999-2001), during which I became a true believer.  Why nott?

Year One's picks were provided by Lipper based on November 30 numbers and drawing from a universe very different from ours.  We got lucky when we found Fidelity funds corresponding to all three of Lipper's picks, but our original formula did not pick any of them.

Year Two was the first year I applied a formula directly to Fidelity funds.  Like Lipper the year before, however, I used November 30 numbers.  I then made a faulty presumption:  that large cash outflows would correspond to big declines in share prices.  I was making the calculations longhand back then, so instead of calculating the cash flows of all of the funds in our universe, I only looked at the dozen or so worst-performing funds.  Was I lazy, stupid, or both?  You tell me…

I have since learned that the funds with the greatest cash outflows are not always found among the poorest performers.  Again, when the back-test used December 31 numbers and drew on our entire universe of funds, our formula did not pick any of Year Two's or Year Three's actual trades.  The purpose of the back-test was to test the formula, so I excluded any funds the formula did not pick.  Too bad, because in our first three years of trading SweetSpot (such as it was), here’s how it did:

Average return per position-year:  +16.4%

Average market return during same period:  -2.5%

Average excess return per position-year:  18.9 percentage points

If we added those results to our baseline numbers, our margin of victory over the market would increase from three percentage points to five.

Starting in Year Four (2002), I used December 31 numbers and calculated cash flows for all funds in our universe.  Our picks from that point to the present match those generated by the back-test (except for my accidental discovery of a second formula).

[2] Three-Year Trades Broken Down:

Year One:

Baseline average return per position-year:  11.1%

Baseline average market return during same period:  12.9%

Baseline average excess return per position-year:  -1.8 points

Consolidated average return per position-year:  11.6%

Consolidated average market return during same period:  12.1%

Consolidated average excess return per position-year:  -0.5 points

Year Two:

Baseline average return per position-year:  16.4%

Baseline average market return during same period:  11.8%

Baseline average excess return per position-year:  4.6 points

Consolidated average return per position-year:  16.4% 

Consolidated average market return during same period:  11.9%

Consolidated average excess return per position-year:  4.5 points

Year Three:

Baseline average return per position-year:  19.4%

Baseline average market return during same period:  12.8%

Baseline average excess return per position-year:  6.6 points

Consolidated average return per position-year:  19.9%

Consolidated average market return during same period:  13.1%

Consolidated average excess return per position-year:  6.9 points