Trading Roofs

The idea behind pattern pairs is to pick a chart pattern type (like broadening bottoms with upward breakouts) to buy and another to sell (like double tops). You buy the upward breakout from the broadening bottom, hold for a few years, and sell when a double top appears and breaks out downward. Along the way, you give price a chance to rise far enough to overcome those trades which are stopped out for a loss. This is a trend-following strategy.

Trading Roofs: Summary

The figure illustrates the idea for trading pattern pairs, where price is the red line and the boxes are chart patterns. This articles assumes you buy an upward breakout from either a roof or a busted one (price breaks out downward, drops no more than 10%, reverses, and closes above the top of the pattern. Buy as price rises above the top of the pattern).

On the sale side, you can sell the first bearish chart pattern which comes along or wait for your favorite bearish chart pattern to appear and sell then.

Here's a list of the top five performing sell signals, based on annualized gain (annualized because the hold time is often years, in parenthesis).

The following list shows the expected performance of chart pattern pairs, ranked by their expectancy. Expectancy is a way of gauging winning and losing trades and how much money you might make trading a pattern pair. I put the expected profit per trade, per share, in parenthesis.

Trading Roofs: Entry and Exit Conditions

The databases I built over several decades doesn't identify every chart pattern. There may be plenty of double tops over the years, for example, that I didn't catalog on the way to the one I did catalog. So buying an upward breakout from a roof and selling at the double top I cataloged would be different than choosing to sell a different double top. However, the following analysis does give a real-world flavor for how well you might do trading chart patterns if you follow the pattern pair strategy.

I used the following 43 chart patterns in the analysis, but some only applied if they were busted.

Trading Roofs: Stops

I used a stop loss order set a penny below the bottom of the chart pattern. Price on the way down may have gapped below the stop price (for the sale price), so I used the lower of the stop price or the opening price on the day of sale).

For trailing stops, I removed the stop loss order and used a trailing stop set at 10%, 15%, 20%, or 25% below a peak, never lowering the stop value, but raising it if a higher peak came along during the trade.

In Table 1, I calculated the percentage net gain (the average of gains and losses) when using various trailing stop loss amounts (10%, 15%, 20%, and 25%) for all tested chart patterns according to the busted/non-busted buy/sell configuration. In parenthesis is the size of the average loss so I could detail how losses change with various stop loss orders.

For example, if I tested non-busted roofs and sold various non-busted patterns (ascending triangles, broadening tops, head-and-shoulders tops, and so on), I made an average of 26% ("Stop Loss Only" column) after having a stop loss order in place. Losses averaged 7%. Replacing the stop loss with a 10% trailing stop cut the gain to 7% but also trimmed the average loss to 5%. Using a 25% trailing stop allowed me to keep more money, 21%, but losses climbed to 11%. If I didn't use any type of stop, the gain averaged 109% with losses averaging 30%.

The results show that trades which do not use a stop make the most money but losses are huge. In four of six columns, non-busted patterns (buys) outperformed busted ones.

Table 1: Various Trailing Stop Settings: Net Profit and (Average Loss)
Data	10%	15%	20%	25%	Stop Loss Only	No Stop
Non-busted buys, non-busted sales	7% (-5%)	16% (-7%)	21% (-10%)	21% (-11%)	26% (-7%)	109% (-30%)
Busted buys, non-busted sales	6% (-5%)	9% (-6%)	18% (-9%)	23% (-15%)	28% (-9%)	89% (-33%)
Non-busted buys, busted sales	5% (-7%)	14% (-8%)	19% (-9%)	20% (-11%)	30% (-7%)	133% (-28%)
Busted buys, busted sales	7% (-5%)	8% (-7)	16% (-9%)	15% (-15%)	16% (-8%)	115% (-33%)

Trading Roofs: Busted Patterns

Table 2 shows what I found when comparing the performance of non-busted patterns (both buy and sell) with busted and non-busted chart patterns. In 22 or 31 contests (up to 31 different chart pattern types, depending on which apply), I compared the three combinations of busted and non-busted buy and sell signals to non-busted buy and sell signals. The table below shows the percentage of time the busted combination beat the non-busted combination in the contests.

For example, I found that busted patterns won up to 39% of the time or 12 of 31 contests when trading (buying and selling) using a busted buy but non-busted sale pattern (compared to non-busted for both).

Trading Roofs: Non-busted Buy, Non-Busted Sale

Table 2: Busted or Non-Busted Contest Winners
	Sell Non-Busted Pattern	Sell Busted Pattern
Buy non-busted pattern	N/A (benchmark)	32%
Buy busted pattern	39%	36%

Table 3 shows statistics I collected for roofs using the trading rules described above and shown in the figure. A stop loss order was used and priced a penny below the bottom of the chart pattern (after buying).

For example, if you were to buy the upward breakout from a roof chart pattern and hold it until you encountered a broadening bottom (the first chart pattern listed in the table), but one with a downward breakout, you'd net an average of 8% on the 33 (3 winners, 31 losers) trades. That's an average of 185% on your winners, 9% average loss on your losers. You'd find that only 9% of the trades made money but you'd gain an average of 9% per year (ranking 31st where 1 is best). If you removed the stop loss order and just held on until the broadening bottom with a downward breakout appeared, you'd make an average of 52% per trade.

The expectancy averages a loss of $0.37 per share per trade which ranks 42nd where 1 is the best value. It suggests you'll have a hard time making money by trading this pair.

Notes: The rank is based on the net gain for all four performance tables (tables 3 to 6) shown below. Trades with sample counts below 30 are not ranked.

Table 3: Statistics for Roofs
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Broadening bottom	185%	-9%	8%	9%	31	52%	3/31	9%	-$0.37	42
Broadening top	156%	-7%	29%	15%	22	98%	15/54	22%	$1.87	28
Broadening formation, right-angled and ascending	117%	-8%	17%	11%	29	61%	7/29	19%	$0.49	37
Broadening formation, right-angled and descending	801%	-8%	63%			120%	2/21	9%
Broadening wedge, ascending	161%	-6%	39%			131%	6/16	27%
Broadening wedge, descending	329%	-5%	36%			71%	3/21	13%
Bump-and-run reversal top	85%	-6%	13%	11%	27	78%	18/68	21%	$1.44	32
Diamond bottom	88%	-6%	32%			94%	2/3	40%
Diamond top	45%	-6%	2%			63%	4/22	15%
Adam & Adam double top	273%	-6%	36%	28%	9	195%	23/132	15%	$4.83	9
Adam & Eve double top	103%	-6%	4%	4%	41	137%	7/69	9%	$2.03	24
Eve & Adam double top	95%	-6%	14%	10%	30	82%	18/71	20%	$1.96	26
Eve & Eve double top	445%	-7%	58%	34%	5	141%	11/65	14%	$2.72	19
Falling wedge	29%	-8%	-5%			29%	2/18	10%
Head-and-shoulders top	266%	-6%	38%	27%	10	135%	39/199	16%	$2.92	17
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Head-and-shoulders, complex top	64%	-7%	8%	8%	34	66%	8/30	21%	$1.70	29
Rectangle top	141%	-8%	20%	13%	25	77%	9/40	18%	$0.97	35
Rising wedge	217%	-7%	41%	25%	12	117%	15/55	21%	$3.35	14
Rounding top	48%	-6%	7%	5%	38	76%	14/44	24%	$2.67	21
Ascending scallop	933%	-9%	36%			48%	1/20	5%
Descending scallop	119%	-6%	13%	9%	32	73%	20/113	15%	$0.20	40
Scallop, inverted and ascending	None	-7%	-7%			88%	0/15	0%
Scallop, descending and inverted	233%	-8%	39%	20%	17	89%	19/78	20%	$2.33	23
Triangle, ascending	49%	-6%	8%	6%	37	72%	11/32	26%	$1.70	30
Triangle, descending	98%	-6%	1%	1%	44	88%	3/38	7%	$3.47	13
Triangle, symmetrical	85%	-6%	6%	5%	39	102%	15/96	14%	$1.03	34
Triple top	254%	-7%	41%	25%	13	115%	49/217	18%	$1.87	27
Rectangle bottom	7%	-7%	-6%	-9%	45	11%	2/38	5%	-$1.74	45
3 falling peaks	182%	-7%	27%	17%	21	130%	42/193	18%	$1.38	33
Roof	23%	-5%	1%			13%	3/10	23%
Roof, inverted	902%	-6%	89%			285%	2/17	11%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank

Trading Roofs: Busted Buy, Non-Busted Sale

The figure shows the setup for this scenario. When price busts the bearish chart pattern (busts a downward breakout from a roof in a bull market), buy. Sell after a downward breakout from the target chart pattern.

Table 4 shows the performance of busted roofs for the entry and non-busted sales after downward breakouts from various bearish chart patterns. A stop loss order was used and priced a penny below the bottom of the pattern (after buying).

A busted roof has a downward breakout but price drops no more than 10% before reversing and moving above the top of the pattern. Buy when price moves at least a penny above the top of the busted pattern. Sell after price drops at least a penny below the target chart pattern.

For example, buying a roof with a busted downward breakout in a bull market (the entry price is really the higher of a penny above the top of the pattern or the opening price) and selling after the downward breakout from a broadening top shows winners averaging gains of 200%. Losses average 9%, for a net of 3%. Only 39 trades occurred with a win/loss ratio of 5%. This scenario ranks the annualized net gain as 42 among the four tables. If you traded this as a buy-and-hold position, meaning no stops were used, the net gain climbed to 89%. Expectancy was a loss of $0.57 per share, ranking 43rd where 1 is best.

Table 4: Statistics for Busted Buys, Normal Sales
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Broadening bottom	8%	-8%	-4%			17%	4/13	24%
Broadening top	200%	-9%	2%	3%	42	89%	2/37	5%	-$0.57	43
Broadening formation, right-angled and ascending	182%	-12%	39%			67%	5/14	26%
Broadening formation, right-angled and descending	122%	-9%	27%			62%	3/8	27%
Broadening wedge, ascending	48%	-9%	6%			64%	5/13	28%
Broadening wedge, descending	299%	-16%	10%			139%	1/11	8%
Bump-and-run reversal top	202%	-10%	45%	32%	6	96%	15/42	26%	$4.43	11
Diamond bottom	6%	-12%	-9%			36%	1/5	17%
Diamond top	18%	-10%	-7%			36%	2/16	11%
Adam & Adam double top	223%	-8%	46%	22%	14	156%	24/79	23%	$6.70	4
Adam & Eve double top	121%	-9%	37%	14%	23	148%	14/26	35%	$5.90	6
Eve & Adam double top	127%	-8%	33%	18%	18	111%	14/32	30%	$8.27	2
Eve & Eve double top	184%	-9%	38%	26%	11	92%	10/31	24%	$5.32	8
Falling wedge	194%	-11%	9%			39%	1/9	10%
Head-and-shoulders top	186%	-9%	35%	22%	15	101%	34/116	23%	$4.44	10
Head-and-shoulders, complex top	17%	-10%	-6%			21%	3/18	14%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Rectangle top	38%	-7%	-1%			20%	3/20	13%
Rising wedge	108%	-9%	18%	13%	24	39%	12/40	23%	$2.00	25
Rounding top	75%	-11%	4%			38%	5/23	18%
Ascending scallop	1%	-10%	-8%			12%	1/5	17%
Descending scallop	172%	-12%	39%	21%	16	60%	24/62	28%	$3.68	12
Scallop, inverted and ascending	None	-6%	-6%			26%	0/3	0%
Scallop, descending and inverted	143%	-13%	15%	13%	26	37%	10/47	18%	-$1.04	44
Triangle, ascending	33%	-6%	0%			81%	3/17	15%
Triangle, descending	21%	-7%	-1%			26%	4/14	22%
Triangle, symmetrical	80%	-9%	5%	5%	40	121%	10/55	15%	-$0.15	41
Triple top	201%	-9%	48%	29%	7	139%	33/89	27%	$5.73	7
Rectangle bottom	78%	-12%	0%			-2%	2/13	13%
3 falling peaks	215%	-9%	31%	18%	19	85%	24/111	18%	$2.40	22
Roof	26%	-7%	4%			43%	3/6	33%
Roof, inverted	234%	-6%	68%			94%	4/9	31%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank

Trading Roofs: Non-busted Buy, Busted Sale

A bullish chart pattern appears and you buy at the breakout. Continue holding until your selected chart pattern appears. The chart pattern is bullish because it has an upward breakout but then things go wrong. Price reverses. Sell when the stock dips below the bottom of the chart pattern (meaning it busts the upward breakout).

Table 5 shows the performance statistics for this setup (buying a normal roof and selling only after a busted chart pattern appears). A stop loss order was used and priced a penny below the bottom of the chart pattern (after buying).

For example, buying a roof with an upward breakout in a bull market and selling a busted broadening top shows winning trades making an average of 306%. Losing trades lost 4% giving a net gain of 16%. The annualized gain is 17% in this case, giving the setup a rank of 20th (where 1 is best). If you traded this without a stop, the net gain climbed to 143%. Of the stocks I looked at, I found 30 trades with 7% of them winning. Expectancy was a gain of $3.13 per share, ranking 15th where 1 is best.

Table 5: Statistics for Normal Buy, Busted Sale
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Broadening bottom	64%	-8%	5%			45%	3/13	19%
Broadening top	306%	-4%	16%	17%	20	143%	2/28	7%	$3.13	15
Broadening formation, right-angled and ascending	None	-9%	-9%			53%	0/13	0%
Broadening formation, right-angled and descending	86%	-5%	5%			68%	1/8	11%
Broadening wedge, ascending	142%	-4%	45%			71%	2/4	33%
Broadening wedge, descending	62%	-7%	3%			99%	2/11	15%
Bump-and-run reversal bottom	294%	-4%	14%			52%	1/15	6%
Cup with handle	None	-9%	-9%			51%	0/4	0%
Diamond bottom	None	-8%	-8%			-8%	0/7	0%
Diamond top	197%	-7%	49%			154%	3/8	27%
Adam & Adam double bottom	132%	-5%	10%	8%	35	161%	6/50	11%	$6.59	5
Adam & Eve double bottom	1049%	-8%	129%	57%	4	320%	4/27	13%	$9.39	1
Eve & Adam double bottom	1326%	-6%	127%	80%	1	264%	3/27	10%	$3.10	16
Eve & Eve double bottom	38%	-9%	3%			75%	5/14	26%
Falling wedge	29%	-7%	-2%			196%	2/14	13%
Head-and-shoulders bottom	51%	-6%	3%	2%	43	85%	8/44	15%	$0.63	36
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Head-and-shoulders complex bottom	None	-7%	-7%			106%	0/7	0%
Rectangle top	91%	-6%	-2%			81%	1/22	4%
Rising wedge	None	-10%	-10%			225%	0/6	0%
Round bottom	52%	-11%	1%			-1%	1/4	20%
Rounding top	None	-5%	-5%			28%	0/5	0%
Ascending scallop	78%	-6%	0%			82%	1/12	8%
Descending scallop	34%	-8%	1%			49%	3/11	21%
Scallop, inverted and ascending	37%	-9%	3%			82%	6/17	26%
Scallop, descending and inverted	417%	-13%	30%			51%	1/9	10%
Triangle, ascending	76%	-6%	6%			93%	2/11	15%
Triangle, descending	118%	-8%	23%			105%	4/12	25%
Triangle, symmetrical	667%	-6%	97%	57%	3	201%	9/50	15%	$2.71	20
Triple bottom	279%	-8%	41%	28%	8	167%	11/54	17%	$1.49	31
Rectangle bottom	None	-6%	-6%			63%	0/10	0%
3 rising valleys	113%	-5%	16%	11%	28	109%	8/38	17%	$0.24	39
Roof	92%	-9%	0%			107%	1/10	9%
Roof, inverted	197%	-5%	62%			104%	2/4	33%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank

Trading Roofs: Busted Buy, Busted Sale

Table 6 shows the last combination of trading statistics. It shows busted roofs as the entry signal and various busted chart patterns as the exit signal. A stop loss order was used and priced a penny below the bottom of the chart pattern (after buying).

For example, buying a busted roof and selling a busted Adam & Adam double bottom made 108% from the winners, lost 10% on the losers for a net gain of 10%. Annualized, it was 7%. The net gain placed the performance of this setup at 36, where 1 is best. Removing stops from the trades allowed them to make 208%. Only 36 trades were taken and 17% of them were winners. Expectancy was a gain of $0.38 per share, ranking 38th among the four tables.

Table 6: Statistics for Busted Buys and Sales
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Broadening bottom	12%	-10%	-7%			185%	1/7	13%
Broadening top	66%	-7%	7%			77%	3/12	20%
Broadening formation, right-angled and ascending	None	-13%	-13%			-44%	0/4	0%
Broadening formation, right-angled and descending	88%	-12%	13%			22%	1/3	25%
Broadening wedge, ascending	87%	-9%	15%			35%	1/3	25%
Broadening wedge, descending	None	-7%	-7%			-2%	0/6	0%
Bump-and-run reversal bottom	None	-6%	-6%			172%	0/9	0%
Cup with handle	130%	-14%	34%			58%	2/4	33%
Diamond bottom	118%	-6%	8%			405%	1/8	11%
Diamond top	86%	-8%	4%			79%	1/7	13%
Adam & Adam double bottom	108%	-10%	10%	7%	36	208%	6/30	17%	$0.38	38
Adam & Eve double bottom	103%	-13%	14%			114%	3/10	23%
Eve & Adam double bottom	40%	-6%	4%			150%	3/11	21%
Eve & Eve double bottom	42%	-8%	1%			10%	2/9	18%
Falling wedge	17%	-6%	0%			25%	1/3	25%
Head-and-shoulders complex bottom	None	-4%	-4%			29%	0/1	0%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank
Rectangle top	55%	-7%	22%			89%	7/8	47%
Rising wedge	None	-4%	-4%			48%	0/4	0%
Round bottom	None	-19%	-19%			89%	0/1	0%
Rounding top	106%	-16%	25%			54%	1/2	33%
Ascending scallop	None	-14%	-14%			120%	0/5	0%
Descending scallop	None	-9%	-9%			105%	0/5	0%
Scallop, inverted and ascending	12%	-9%	-7%			45%	1/10	9%
Scallop, descending and inverted	129%	-10%	30%			101%	2/5	29%
Triangle, ascending	13%	-7%	-2%			-6%	3/9	25%
Triangle, descending	72%	-9%	9%			21%	3/11	21%
Triangle, symmetrical	355%	-6%	74%	59%	2	130%	10/35	22%	$7.82	3
Triple bottom	97%	-7%	21%	9%	33	100%	9/25	26%	$2.87	18
Rectangle bottom	None	-5%	-5%			481%	0/5	0%
3 rising valleys	170%	-11%	15%			122%	3/18	14%
Roof	39%	-6%	3%			105%	1/4	20%
Roof, inverted	25%	-7%	-1%			83%	1/5	17%
Sell Pattern	Average Win	Average Loss	Net	Annualized Net	Rank	No Stop Net	Win/Loss Samples	Win Loss	Average Expectancy	Expectancy Rank

Trading Roofs: Performance Improvements

Here are a few ideas the data suggested which may improve performance of your pattern pairs trading.

Trend Start: Short, Medium, or Long

Find the trend start for your chart pattern. Often you can just look at a chart and see where the trend begins. If not, or you want to be sure, then the glossary describes how to find it.

Determine the length from the trend start to the pattern's start: short term (less than 3 months), medium term (3 to 6 months) or long term (more than 6 months).

Table 7 shows the results for the four combinations of busted/non-busted trades and the resulting performance.

Buying non-busted patterns with a short-term (up to 3 months) duration from the trend start to the pattern's start results in better performance. Busted patterns do best using a medium-term (3 to 6 months) duration from the trend start.

Moving Averages: 50- and 200-Day SMA

Table 7: Short (S) Medium (M) or Long (L) Trend Start and Performance
	Sell Non-Busted Pattern	Sell Busted Pattern
Buy non-busted pattern	S49% M6% L-1%	S57% M10% L-1%
Buy busted pattern	S24% M32% L28%	S9% M28% L15%

I checked two moving averages at buy time, 50- and 200-day simple moving averages (not as a crossover setup). I compared the breakout price to the value of the moving average. Table 8 shows the performance of buying or selling busted or non-busted patterns when the breakout price was above (A) or below (B) the 50-day simple moving average (SMA).

In all cases, buying if the breakout price is above the 50-day SMA gives the best results.

Table 8: Above (A) Below (B) 50-Day Simple Moving Average
	Sell Non-Busted Pattern	Sell Busted Pattern
Buy non-busted pattern	A31% B6%	A36% B5%
Buy busted pattern	A31% B-5%	A19% B-8%

Table 9 shows the results of using a longer moving average, the 200-day. Traders often use this as a proxy for the long-term trend.

Buying roof patterns when the breakout price is above the 200-day SMA gives the best results.

Selling First Bearish Chart Pattern

The prior discussion assumes you buy a roof (busted or non-busted) but sell a chart pattern of your choosing, such as a downward breakout from a head-and-shoulders top (you wait for one to appear). What if you sold the first bearish chart pattern which comes along? How would you do?

Table 9: Above (A) Below (B) 200-Day Simple Moving Average
	Sell Non-Busted Pattern	Sell Busted Pattern
Buy non-busted pattern	A31% B9%	A35% B8%
Buy busted pattern	A32% B-3%	A19% B-1%

Table 10 shows the results sorted by the type of patterns involved (busted or non-busted). For example, if you buy a non-busted roof and sell the first non-busted chart pattern which comes along, you'd make 8% on average. Annualized, you'd make 27%. This compares to a 18% annualized gain if you sell a designated pattern (like you waited for a double top before selling, which may or may not be the first bearish chart pattern to come along).

The results split into non-busted and busted sales. Selling the first non-busted pattern which appears or selling a designated busted pattern worked best. Note that selling the first non-busted pattern will make you more money than selling a designated busted pattern (on average).

The bottom half of the table shows expectancy for the four combinations. Buying and selling busted patterns gives the highest expectancy ($3.42).

Bulkowski on Pattern Pairs: Roofs