Nov

27

A Post, from Victor Niederhoffer

November 27, 2012 |

A post purporting to show that buy and hold investing does not work has appeared on our list. It is reprehensible propaganda and total mumbo. They do not take account of the distribution of returns to investing over long periods that have been enumerated by the Dimson group and Fisher and Lorie. It is sad to see this on our site. The arguments against buy and hold seem to be that the professors found that short term investing didn't work so they erroneously concluded that long term investing must be the alternative. Shiller is mentioned and cited with approval.

Alston Mabry writes: 

To explore this issue numerically, I took the monthly data for SPY (1993-present) and compared some simple fixed systems. In each system the investor is getting $1000 per month to invest. If during that month, the SPY falls a set % below the highest price set during a specific lookback period (the 3, 6, 12, 18, 24 or 36 months previous to the current month), then the investor buys SPY with all his current cash (fractional shares allowed). If the SPY does not hit the target buy point this month, then the $1000 is added to cash. Once the investor buys SPY shares, he holds them until the present.

For example, let's say the drop % is 10%, and the lookback period is 12 months. In May of year X, we look at the high for SPY from May, year X-1, thru April, year X, and find that it is 70. We're looking for a 10% drop, so our target price would be 63. If we hit it, then spend all available cash to buy SPY @ 63. Otherwise we add $1000 to cash.

Each combination of % drop and lookback period is a separate fixed system.

Over the time period studied, if the investor just socks away the cash and never buys a share (and earns no interest), he winds up with $239,000. On the other hand, if he never keeps cash but instead buys as much SPY each month as he can for $1000, then he winds up with over $446,000, which amount I use as the buy-and-hold benchmark.

If the investor uses the fixed system described, he winds up with some other amount. The table of results shows how each combination of % drop and lookback period compared to the benchmark $446,000, expressed as a decimal, e.g., 0.78 would that particular combination produced (0.78 * 446000 ) dollars.

Results in this table
.

The best system was { 57% drop, 18+ month lookback }, or just to wait from 1993 until March 2009 to buy in. Of course, it's hard to know that 57% ex ante. The next best system was { 7% drop, 3 month lookback } coming in at 0.99.

This study is just food for thought. It leaves out options for investing cash while not in the market. And it sticks with fixed %'s without exploring using standard deviation of realized volatility as a measure. So, there are other ways to play with it.

Charles Pennington comments: 

Thank you — that is a remarkable "nail-in-the-coffin" result.

Nothing beat buy-and-hold except for the ones with the freakish 57% threshold, and it won by a tiny margin, and it must have been dominated by a few rare events–57% declines–and therefore must have a lot of statistical uncertainty..

That's very surprising and very convincing.

(Now some wise-guy is going to ask what happens if you wait until the market is UP x% over the past N months rather than down!)

Kim Zussman writes: 

Here are the mean monthly returns of SPY (93-present) for all months, months after last month was down, and months after last month was up (compared to mean of zero):

 One-Sample T: ALL mo, aft DN mo, aft UP mo

Test of mu = 0 vs not = 0

Variable      N      Mean     StDev   SE Mean  95% CI            T
ALL mo     237  0.0073  0.0437  0.0028  ( 0.0017, 0.0129)  2.58
aft DN mo   90   0.0050  0.0515  0.0054  (-0.0057, 0.0158)  0.92
aft UP mo  146  0.0083  0.0380  0.0031  ( 0.0021, 0.0145)  2.65

 The means of all months and months after up months were significantly different from zero; months after down months were not.

Comparing months after down vs months after up, the difference is N.S.:

Two-sample T for aft DN mo vs aft UP mo

                  N    Mean   StDev  SE Mean
aft DN mo   90  0.0050  0.0515   0.0054   T=-0.53
aft UP mo  146  0.0084  0.0381   0.0032

Bill Rafter writes: 

A few years ago I published a short piece illustrating research on Buy & Hold. It contrasted a perfect knowledge B&H with a variation using less-than-perfect knowledge using more frequent turnover. Here's the method, which can easily be replicated:

Pick a period (say a year) and give yourself perfect look-ahead bias, akin to having the newspaper one year in the future. Identify those stocks (say 100) that perform best over that period, and simulate buying them. Over that year you cannot do better. That's your benchmark.

Then over that same period do the following: Buy those same 100 stocks, but sell them half-way thru the period. Replace them at the 6-month mark with the 100 stocks perfectly forecast over the next 12 months. Again sell them after holding them for just half the period. Thus the return from the stocks that you have owned and rotated are the result of less-than-perfect knowledge. Compare that return to the benchmark.

Do this every day to eliminate start-date bias, and then average all returns. The less-than-perfect knowledge results far exceeded the perfect-knowledge B&H. Actually they blew them away in every time frame. It's really obvious when you do this with monthly and quarterly periods as you have so many of them.

The funny thing about this is the barrage of hate mail that I received from dedicated B&H investment advisors, who somehow felt their future livelihoods were threatened.

If anyone wants that old article, send me a message off the list. We called it "Cassandra" after someone with perfect knowledge that was scorned.

Anton Johnson writes in: 

Here is a link to BR's excellent study "Cassandra", as it lives on in cyberspace.


Comments

Name

Email

Website

Speak your mind

2 Comments so far

  1. Ed on November 27, 2012 2:03 pm

    Alson, I have found that using historically accurate rates to simulate return on cash can radically alter the performance of timing systems over long periods of time. It is easy to underestimate this from today’s vantage-point of an ultra-low interest rate environment.

  2. Cole Walton on November 27, 2012 4:50 pm

    Does the Cassandra study take into account taxes? Like most things in life, I suspect there is a “happy medium” between active vs. passive investment. Also, how do you measure the threshold a new piece of information must pass for it to be actionable? How do you filter out noise vs. relevant news items? This study seems to be way too simplistic as it ignores many real world variables such as who is doing the trading :)

Archives

Resources & Links

Search