Identifying and evaluating horizontal support and resistance levels: an empirical study on US stock markets

Abstract

We propose a novel rule-based mechanism that identifies Horizontal Support And Resistance (HSAR) levels. The novelty of this system resides in the manner it encloses principles, found in well known technical manuals, used for the identification via visual assessment. The drawing of these levels derives from historical locals, rather than denoting support (resistance) levels from the lowest (highest) price levels of precedent constant time intervals. We further proceed in evaluating whether these levels are efficient trend-reversal predictors, and if they can generate systematic abnormal returns. The dataset used includes adjusted for dividends and splits, daily closing prices of stocks listed on National Association of Securities Dealers Automated Quotation (NASDAQ) and New York Stock Exchange (NYSE) for the last 2 decades. Our results are aligned with the efficient market hypothesis. More concretely, support levels outperform resistance ones in predicting trend interruptions but they fail to generate excess returns when they are compared with simple buy-and-hold strategies.

Keywords:

• technical analysis
• support and resistance levels
• efficient market hypothesis
• stock markets
• trading strategies

JEL Classification::

• G14
• G15
• C60

Acknowledgements

We would like to thank an anonymous referee of this journal for his helpful comments.

Notes

¹ For example, the neckline of the head-and-shoulders pattern is considered as support level before its penetration and as a resistance afterwards.

² In these studies this type of SAR levels is being referred also as trading range break-outs.

³ Commercial banks, investment banks and real-time information providers are included in these firms. The author did not provide the firms’ names to preserve anonymity.

⁴ The corresponding formulas for identifying this type of SAR levels can be found in the Appendix.

⁵ Functions presented in this article have the following general form: [output₁, output₂, … , output _n ] = function's name (input₁, input₂, … , input _n ). The variables in the squared brackets are the outputs generated by the corresponding function and the variables inside the parenthesis are the necessary inputs.

⁶ For example, when observing the frequencies of peaks and bottoms between two bins (b_a and b_b ), whose bounds are 10 to 11 and 20 to 21 respectively; the corresponding percentages would be 10% and 5% respectively. In other words in the first case a SAR zone is defined which includes locals that has distance greater than those included in the second one.

⁷ The first 2 years are used for the preliminary estimation of SAR for the first sub-period and for the whole sample period. The second and third sub-periods of comparison used for the same purpose all previous, available information.

⁸ 45 000 results by creating 10 artificial SAR, for each of the 250 trading days per year, for 18 years.

⁹ We also used rolling windows of 50 and 150 days and desired distances of 2%, 4% and 5%.

¹⁰ If the number of trials ‘n’ is great enough and neither p (probability of positive outcome) nor q (probability of negative outcome) is close to zero then a standard normal distribution can approximate a binomial distribution sufficiently well with a variable: , where X is the number of successful trials. In practice the approximation is very good if np and nq are greater than 5 (Murray, 1975).

¹¹ The results for NYSE are available from the author upon request.

¹² The results reported in this study are robust, and not affected with a use of arithmetic returns. The choice of logarithmic returns was to enhance the code performance by taking the benefit of logarithmic returns’ properties. For example, sums of logarithmic returns over a time interval, gives the logarithmic return for that interval.

¹³ The corresponding graphs are available upon request.

¹⁴ For the examined period, other studies also report results that are in favour of the Efficient-Market Hypothesis (EMH), even though they use different technical indicators. For example, Milionis and Papanagiotou (2009, 2011) found no significant predictive power of the moving average trading rule when applied on the SP-500 index of NYSE for the period 1993–2005. Kwon and Kish (2002) tested the efficacy of a variety of trading rules (the simple price moving average, the momentum and trading volume) and they found that in the last sub-period they examined (1985–1996) the NYSE index is getting more efficient in information due to technological improvements.