10% owner at 4 companies we cover.
14 open-market purchases and 3 sales not made under a pre-set plan, reported from April 2, 2010 to August 27, 2020.
Would copying Douglas Family Trust’s purchases have beaten the S&P 500?
Too few purchases to judge. Only 12 purchases have 12 months of prices after the report; we judge insiders once they have 20. The figures below can swing a long way on a handful of trades. Buying each purchase the day after it was reported, an equal amount each, and holding it 12 months or until they reported a sale would have turned $10,000 on April 5, 2010 into $3,989 by November 2, 2026, −5.4% a year. The copy held shares only 30% of the time, so the fair comparison is the S&P 500 on those same days with Treasury bills otherwise: $22,538, +5.0% a year (the S&P 500 held throughout made $85,213, +13.8% a year). Beta against the index was 0.41. Twelve months after each report, the average purchase was −29.2% against the S&P 500 (median −20.8%), and 33% beat it. Measured from the trade date, the insider’s own timing, the average was −32.7%.
With 296 insiders tested, some will look skilled or unskilled by chance alone: when we replaced each insider’s stocks with random companies bought on the same days, about 5.7 cleared the bar ahead and 23.3 behind. One record on its own proves little.
Copying Douglas Family TrustS&P 500, same days investedS&P 500, held throughout$10,000 copying Douglas Family Trust’s open-market purchases from April 5, 2010, before costs and taxes, marked weekly. Each purchase gets the same starting stake, bought at the close of the first trading day after the report and sold 12 months later or the day after a reported sale of that stock; uninvested money earns the Treasury bill rate. “Same days invested” holds the S&P 500 only on the days the copy held shares, Treasury bills otherwise.
Return a year, copying
−5.4%
S&P 500, same days invested
+5.0% (held throughout +13.8%)
Time in shares
30%
Sharpe ratio (S&P 500)
−0.07 (0.75)
Biggest drop (S&P 500)
−87% (−34%)
Market sensitivity (beta)
0.41
Purchases copied
12
Trade by trade
Douglas Family Trust: trades against the S&P 500 afterwards
Measured
Trades counted
Average vs S&P 500
Median
Went the right way
Years
t-statistic
p-value
Evidence
Purchases, 3 months after the report
12
−20.3%
−21.5%
8%
5
−5.59
0.005
Some evidence: behind
Purchases, 6 months after the report
12
−36.6%
−24.4%
8%
5
−3.75
0.020
Some evidence: behind
Purchases, 12 months after the report
12
−29.2%
−20.8%
33%
5
−1.82
0.14
No reliable evidence
Purchases, 12 months after the trade
12
−32.7%
−24.8%
42%
5
−2.09
0.10
No reliable evidence
Sales, 3 months after the report
0
—
—
—
—
—
—
Not enough years
Sales, 12 months after the report
0
—
—
—
—
—
—
Not enough years
Sales, 12 months after the trade
0
—
—
—
—
—
—
Not enough years
“Average vs S&P 500” is the stock’s return including dividends minus the index’s over the same dates, from the close on the first trading day after the report, or from the close on the trade date.
Only windows that have finished are counted. For sales, a negative figure means the stock lagged after it was sold, so selling was the right call (green).
“Went the right way” is the share of purchases that beat the index, or of sales that trailed it.
The t-statistic uses one average per year of reports, because one person’s trades bunch into the same stock and weeks, and needs trades in at least four different years; the p-value comes from the t distribution with that number of years (with four years, a t of 2 has p = 0.14). “Some evidence” means p under 5%, “strong evidence” p under 1% over at least six years.
Open-market purchases and sales of at least $10,000 in companies we cover, from SEC Form 4 filings; sales made under a pre-set Rule 10b5-1 plan are not shown. “Not yet” means the window has not finished or the share price was under $5 at entry.
These are the shares’ moves after each trade: they say nothing about why Douglas Family Trust traded and do not suggest the use of non-public information.
How every insider compares · the insider trading backtest.
