Research results from Investtech, 18 December 2014
Published in English on 30 December 2014. Norwegian original here >>
About the author
Geir Linløkken is the Head of Analysis and Research at Investtech, and is responsible for portfolios and money management. He founded Investtech in 1997, to provide independent technical analyses based on science and investor psychology. Mr. Linløkken has an MSc. in Computer Science, specializing in Mathematical Modeling, at the University of Oslo. He is the author of the book “Technical Stock Analysis”. His daily work includes analysing stocks and developing quantitative methods for stock market investments.
Keywords: head and shoulders formation, inverse head and shoulders formation, buy signal, sell signal, Oslo stock exchange, statistics, technical analysis.
Abstract:
Geometric price patterns, like head and shoulders formations, are used in technical analysis to predict future price development. Many investors use this as an important part of their decision making process when buying or selling stocks. We have looked at the price movements that followed sell signals from head and shoulders formations and buy signals from inverse head and shoulders formations on the Oslo Stock Exchange in a period of 19 years, from 1996 to 2014. Stocks with buy signals have on average increased by 13.9 % after three months, while stocks with sell signals have fallen 4.4 %. Compared to average stock exchange development in the same period, buy signals did 10.7 percentage points better and sell signals did 7.6 percentage points worse.
This research report is part of a bigger research project conducted by Investtech into price development following technical formations in stock prices. This report is on long term head and shoulders formations and inverse head and shoulders formations on the Oslo Stock Exchange in Norway.
Short term | Medium term | Long term | |
Rectangle | Report | Report | Report |
Inverse/ head and shoulders | Report | Report | Present |
Double top, double bottom | Report | Report | Report |
Identification of geometric price patterns in stock prices is an important area of technical analysis. The idea is that these patterns describe the investors’ mental state, i.e. whether they will want to sell or buy stocks in the time ahead, and they thereby indicate the future direction of the stock price. Head and shoulders formations are one type of such patterns.
A head and shoulders formation is a top formation which marks the end of a rising period. The formation consists of a left shoulder, a head and a right shoulder, connected by a neckline, see figure 1. The creation of a head and shoulders formation indicates increasing pessimism among investors and the start of a falling trend. Such formations are considered among the most reliable signals in technical analysis. They are primarily used to predict reversals in long term market trends, but can also be used in the shorter term.
This formation also exists in the opposite direction, as an inverse head and shoulders formation, see figure 2. This is a bottom formation which marks the end of a falling period. An inverse head and shoulders formation signals increasing optimism among investors and the start of a rising trend.
Figure 1: Sell signal from head and shoulders formation.
Figure 2: Buy signal from inverse head and shoulders formation.
In technical analysis terminology we say that a break downwards through the neckline of a head and shoulders formation triggers a sell signal. Similarly a break up from an inverse head and shoulders formation triggers a buy signal. We have studied the price movements following buy and sell signals from such formations on the Oslo Stock Exchange in Norway.
It is no easy task to identify head and shoulders formations in stock prices. The figures above show that the price forms a left shoulder, a head and a right shoulder, before the neckline is broken. However, stock prices are rarely as regular as these illustrations. The shoulders will often be of different sizes, the head and one shoulder may be almost the same size, or the neckline may be crooked.
Many investors identify price patterns by looking at price charts and drawing lines by hand. This method has many weaknesses, most of all that it is subjective, allowing you to see the formations you want to see, and it is very time consuming. Therefore we need an automatic algorithm whereby computers identify the formations and the signals they trigger.
Investtech has studied technical and quantitative analysis since 1997. We have developed mathematical algorithms for automatic identification of head and shoulders formations in stock prices. The formations are entered into the technical analysis charts, shows in signal lists and presented updated daily to Investtech’s subscribers.
In this report we have looked at the price movements that follow buy and sell signals from head and shoulders formations on the Oslo Stock Exchange. The statistics are based on formations automatically recognized by Investtech’s computer programs. No parameter optimization or changes to algorithms have been made during this study. This is an analysis based on the existing historical material.
We have used stock prices from 1 January 1996 to 10 October 2014 as the basis for the statistics. In this period, the main index on the Oslo Exchange rose from 106.9 to 573.6 points, which is 437 % or approximately 9.3 % a year. Compared to the risk free interest rate in this period, this is approximately what can be expected for similar periods of time.
In eight of these 19 years, the exchange rose by over 30 %, while it fell by more than 10 % in five of the years, and varied between minus 10 % to plus 30 % in five of the years. We have had both good and bad periods, and several sideways periods as well, and consider this representative for a normal period of time on the exchange.
All stocks that have been listed in the period are included. Stocks that have been delisted due to for instance mergers, takeovers and bankruptcy are included. However, we only have data for these companies for as long as they were listed. A company which went bankrupt will then have a final trading price which is not zero, which is a weakness in this study. However, this is only the case for a small number of companies. Most companies also fall a lot before they are delisted, so the difference between the price fall from when they were listed and a price fall down to zero will be small.
It is also very rare that new buy signals are generated from head and shoulders formations when a company’s stock price is falling. Therefore it matters very little to the statistics for buy signals. Return from sell signals would however have been a little weaker had we corrected for bankruptcies. Combined it is our opinion that these conditions have minimal impact on the results of this study.
All prices are adjusted for splits, dividend payments, reverse splits, and other corporate capital changes, in order to reflect the actual value development of the stocks.
715 time series are included, of which 597 are stocks with at least 66 days of trading. At the end of the period, approximately 220 stocks were listed on the exchange.
The stock’s daily closing price is used. We have only used prices and turnover figures from the stock’s primary market place. Alternative markets like Chi-X, Bats and Burgundy are excluded.
We have used Investtech’s algorithms for automatic identification of price formations. The algorithms were run on long term charts made up of 1,399 price days, approximately 6 calendar years. We consider the algorithms good at identifying actual inverse/ head and shoulders formations, and they do not classify indistinct patterns as actual formations.
At identification of signals, only data up to the date the signal was triggered were used. The later data were hidden from the algorithm.
All signals identified from head and shoulders and inverse head and shoulders formations are used. Normally each formation only triggers one signal. However, in rare cases they may trigger several signals. This happens if the price following the break reacts back into the formation, creates a modified formation and then breaks out again.
Sometimes one stock can also trigger several signals on the same day. This happens if the algorithms have recognized several formations of different length and height which are broken out from at the same time.
In order to have the data set as representative for the Oslo Stock Exchange as possible, we remove certain signals from the data set:
Our data set now consists of 111 identified buy signals from inverse head and shoulders formations and 144 sell signals from head and shoulders formations in stocks and equity certificates on the Oslo Stock Exchange in the period 1996 to 2014.
Figure 3: Price development after buy and sell signals from inverse/ head and shoulders formations on the Oslo Stock Exchange identified by Investtech’s automatic algorithms in long term price charts. Click the image for bigger version.
The chart shows average price development following buy and sell signals from head and shoulders formations. The signals are triggered on day 0. Only days when the exchange is open are included, so 66 days equal approximately three months. Buy signals are the blue line and sell signals are the red one. The shaded areas are the standard deviation of the calculations. Benchmark index is the black line.
Buy signal | Day 1 | 10 | 22 | 66 | 250 |
Absolute | 0.72 % | 1.10 % | 3.85 % | 13.93 % | 26.23 % |
Benchmark index | 0.05 % | 0.48 % | 1.05 % | 3.18 % | 12.61 % |
Relative, percentage points | 0.68 | 0.62 | 2.79 | 10.61 | 13.55 |
Statistical T-value | 1.42 | 0.70 | 2.12 | 3.02 | 1.97 |
Buy signals from inverse head and shoulders formations identified in Investtech’s long term charts have historically given increasing prices in the following months. On average, stocks that triggered buy signals have risen 13.9 % in the following three months. This is 10.7 percentage points better than index, which rose 3.2 % in an average 3 month period. The figures are considered significantly positive, with a statistical T-value of 3.1 standard deviations against average index development after 66 days.
The increase is quite even in the first three months following the signal. In the first 22 days, equal to approximately a month, the signal stocks rise on average by 0.15 % per day, estimated with least-squares method adapted to the yield curve. In the next 44 days, the stocks rise by an average of 0.19 % per day.
The average figures are calculated from 111 observations. This is a low number, and the uncertainty of the calculations is therefore fairly large. The blue shaded area shows the standard deviation of these estimates. Assuming normal distribution and independent observations, the average price development following buy signals from inverse head and shoulders formations has a 68 % probability of falling within this interval.
Even though the average figures are significantly positive, there is great variation from signal to signal. A count showed that 65 % of the signals gave return which was positive or zero after 66 days, while 35 % gave negative return. As such there is fairly high probability of losing money from one investment based on one buy signal from one inverse head and shoulders formation, even though the average is positive.
Sell signal | Day 1 | 10 | 22 | 66 | 250 |
Absolute | -0.33 % | -1.22 % | -4.31 % | -4.39 % | 6.94 % |
Benchmark index | 0.05 % | 0.48 % | 1.05 % | 3.18 % | 12.61 % |
Relative, percentage points | -0.38 | -1.69 | -5.37 | -7.71 | -5.74 |
Statistical T-value | -1.45 | -2.47 | -4.62 | -3.02 | -0.77 |
Sell signals from head and shoulders formations identified in Investtech’s long term price charts have historically been followed by markedly falling prices in the following weeks. On average stocks with sell signals have fallen 4.4 % the following three months, equal to 7.6 percentage points weaker than average three month index development.
The fall is stronger in the first four to five weeks following the signals, and after that the stocks develop more sideways. A least-squares method adaptation to the yield curve shows a fall of 0.20 % per day the first 22 days and a completely flat development the next 44 days.
The average figures are calculated from 144 observations. This is a low number. Due to the great deviation from average return in the index, the results are still statistically significant assuming normal distribution of independent data. Statistical T-value after 66 days is -3.0.
As is the case with buy signals, there is great variation from signal to signal. In total 62 % of the signals gave negative or zero return after 66 days, while 38 % gave positive return.
The results are based on average return figures. If any individual stock shows extreme effects, like a rise of several hundred per cent, this may strongly impact the average figures. To investigate this, we have calculated how much each stock weighs in the calculation of the average figures.
Figure 4: Weight per stock in the calculation of average price development following buy signals. The 10 most highly weighted stocks are indicated in their own sectors of the pie chart.
Figure 5: Weight per stock in the calculation of average price development following sell signals.
In the case of buy signals, the ten most heavily weighted stocks make up 41 % of the total. Panfish alone makes 13 %, while the other nine are between 2.4 and 4.5 per cent each. The remaining 59 % consists of 74 different stocks. For the sell signals, the most heavily weighted stocks make up 32 % of the total. The remaining 68 % consists of 87 different stocks.
A relatively large part of the average is made up of a small number of stocks. Except for Panfish’s 13 per cent, however, no stocks weigh over five per cent. This shows that the results are likely not due to extreme results in one or a few individual stocks. However, we would like a more extensive data set in order to get more robust figures and lower estimator variance.
In the calculations above we have included signals from stocks with an average daily turnover of at least half a million Norwegian krone on the Oslo Stock Exchange at the time the signal was triggered. Varying the liquidity parameters allows us to investigate if there is a difference in signal strength for smaller and bigger companies.
Drawing the line at five million krone splits the companies into two roughly similar groups. Signals from stocks with turnover below five million come to 114 signals, and those with turnover above five million come to 141 signals.
Figures 6 and 7 show that buy signals from the group of bigger companies have done better than buy signals from the smaller companies. At the same time, sell signals from the bigger companies have been better, i.e. the price development has been weaker, than for the smaller companies.
The results are based on a small number of observations, which means that the variance of the estimators, illustrated by the shaded areas, is great. Results from the Swedish Stock Exchange also indicate that there signal power is greater for the smaller companies. As such we should not place too much importance on this liquidity analysis.
The signals in this study are identified in Investtech’s long term technical charts. Such charts are often used to analyse the exchange with the price development of the next three to 18 months in mind. It is therefore interesting to look at the price development also in periods beyond 66 days following the signal.
Figure 8: Price development from buy and sell signals from inverse head and shoulders formations and head and shoulders formations on the Oslo Stock Exchange. The chart shows price development from 22 days before the signals were triggered till 250 days after. Click the image for bigger version.
The chart shows that the formations have strong signal power for approximately the first hundred days, equalling almost five months, after they triggered buy or sell signals. Stocks with buy signals have a good increase in this period, and do better than average index development, and stocks with sell signals fall a lot at first and then move sideways. After about five months, however, we first see a fairly neutral development, while the signal stocks then reverse some of the early excess or negative excess return.
We also note that the width of the confidence interval increases drastically. This shows great differences between the signal stocks, and that the significance of the results falls over time. This is a natural development which indicates that the psychological conditions in the market described by the long term head and shoulders and inverse head and shoulders formations have almost disappeared after five months.
The chart indicates that long term head and shoulders and inverse head and shoulders formations have good predictive power in the first five months, but that the signal power thereafter is practically zero.
Figure 8 shows price development for 22 days, approximately one month, before the stocks trigger signals.
The chart shows that stocks with buy signal from inverse head and shoulders formation on average have risen over ten per cent the last 22 days up to and including the day the signal was triggered. It can be psychologically difficult to buy a stock which has risen so much in such a short time. This may be part of the reason why the signal works so well. The stock should rise more, based on the news or fundamental conditions that perhaps triggered the signal, but it does not due to investor psychology and human weakness. Over time the investors accept the basis for the price increase and become more positive, which gives a good rise over the next few months.
The same is true for sell signals. The chart shows that stocks with sell signals from head and shoulders formations fall around nine per cent the last 22 days up to and including the day the signal was triggered. Yet they continue to fall or move sideways the next weeks and months. It may be psychologically difficult to sell a stock which has become much “cheaper” in a short period of time. It takes time for the actual state of affairs to be accepted, fundamental analyses to be adjusted and investors to conquer their own psychological resistance to selling on lower prices. Thus the stock’s weak development continues also after the price fall that triggered the signal.
Figure 9. Price development following buy and sell signals from head and shoulders formations on the Stockholm Stock Exchange.
We have conducted the same study on the Stockholm Stock Exchange in Sweden, for the period 1 April 2003 to 10 October 2014. Investtech’s computers identified 152 buy signals and 231 sell signals in this period.
On the Stockholm Exchange, as well, buy signals from inverse head and shoulders formations were followed by rising prices the next three months, and rose more than the exchange in an average three-month period. Stocks with sell signals from head and shoulders formations had quite a sideways development in the following three months, significantly weaker than average index development.
The standard deviation is great, however, and the results are less clear than for the Oslo Exchange. Liquidity-wise the smaller companies had the better signal power, whereas this was the other way around in Oslo. The sell signals from the Stockholm Exchange had good long term predictability, while the results were better in the shorter term for the Oslo Exchange.
This indicates a small data set and that further studies are necessary.
Read the complete research report for the Stockholm Stock Exchange here (in Swedish).
We have studied return from stocks on the Oslo Stock Exchange with breaks through technical head and shoulders formations and inverse head and shoulders formations over a period of 19 years, from 1996 to 2014. Investtech’s automatic algorithms identified a total of 111 buy signals and 144 sell signals from such formations. The buy signals gave an average return of 13.9 % in the following three months, and sell signals gave a return of -4.4 %. Relative to average benchmark index development, the buy signals gave an excess return of 10.7 percentage points, while the sell signals gave a negative excess return of 7.6 percentage points.
The signals demonstrate strong average results. The time period for the study is fairly long, the quality of the data is considered to be good and the algorithms used are entirely automatic and deemed to identify only actual inverse/ head and shoulders formations. Statistical measures suggest a high degree of significance, which indicates that we have found actual connections between signals from price formations and future return, such that trading based on these formations can also give good results in the future.
However, the numbers are based on relatively few observations, and the results from Sweden are not as good. It is therefore necessary to conduct further studies. In particular it would be interesting to look at other price formations that are meant to signal the same thing as head and shoulders formations, i.e. long term trend reversals. It would be especially interesting to look at the price movement in stocks with signals from long term double top and double bottom formations, and it may also be of interest to look at long term rectangle formations. In addition, the medium long term signal power of head and shoulders formations is of interest, as well as results from these formations in other markets. Longer periods of time would also be preferable, for broader base data with more observations.
Keywords: Buy signal,Head and shoulders formation,Inverse head and shoulders formation,Oslo Stock Exchange,Sell signal,statistics.
"Investtech analyserar psykologin i marknaden och ger dig konkreta tradingförslag varje dag."
Investtech garanterar inte fullständigheten eller korrektheten av analyserna. Eventuell exponering utifrån de råd / signaler som framkommer i analyserna görs helt och fullt på den enskilda investerarens räkning och risk. Investtech är inte ansvarig för någon form för förlust, varken direkt eller indirekt, som uppstår som en följd av att ha använt Investtechs analyser. Upplysningar om eventuella intressekonflikter kommer alltid att framgå av investeringsrekommendationen. Ytterligare information om Investtechs analyser finns på infosidan.
Investtech garanterar inte fullständigheten eller korrektheten av analyserna. Eventuell exponering utifrån de råd / signaler som framkommer i analyserna görs helt och fullt på den enskilda investerarens räkning och risk. Investtech är inte ansvarig för någon form för förlust, varken direkt eller indirekt, som uppstår som en följd av att ha använt Investtechs analyser. Upplysningar om eventuella intressekonflikter kommer alltid att framgå av investeringsrekommendationen. Ytterligare information om Investtechs analyser finns på infosidan.