Marabuzo means “shaven head” in Japanese and describes a candlestick which consists of a long body and very small ‘wicks’ at either end. A typical marabuzo is above average length and represents the predominance of one force – either buying or selling – over the other.
In theory the marabuzo is a continuation candle, meaning that a green marabuzo in an up-trend signals higher prices whilst a red marabuzo in a down-trend signals a continuation lower. This article, however, seeks to put to the test the marabuzo’s fabled powers of indicating continuation using data for the EUR/USD currency pair.
In the diargram below I have highlighted the marabuzo bars in blue. There is no hard and fast definition, but for a candlestick to be termed ‘marabuzo’ it must embody the characteristics of long filled- in body and short wick.
Conclusion in summary
In summary, when tested on EUR/USD, the marabuzo was not as strong an indicator of continuation as previously held. It failed to show a statistically significant likelihood of indicating substantial continuation when back-tested over 13 years of daily data.
When used in conjunction with MACD – to filter out contra-trend instances – the results did improve although not sufficiently to show significance. However, it would not be untrue to say that in a trending market, with the benefit of a wider stop, marabuzo’s which occurred in the same direction as the trend did tend to indicate substantial continuation more times than they did not.
1st Experiment – continuation with price action alone
The first experiment looked at the basic marabuzo bar in isolation without other indicators. I defined marabuzo bar as follows:
a) It had to be longer than the average range of the last 8 bars
b) It had to have wicks which were no greater than 10% of the total range at either end.
Note that despite the 10% or less requirement for wick-length perhaps sounding a little too much, when I actually tried out different values I found it still retained a sufficiently marabuzo-like look to satisfy classification. In addition shorter wick sizes, such as 5% of the range, for example, led to so few examples that the sample was of no consequence.
Next the data was tested to determine if the marabuzo bars really did indicate continuation or not.
I defined continuation as the market extending substantially and immediately after the posting of the marabuzo. In other words and to be more precise I defined continuation as the market extending the full candle-length of the marabuzo again in the direction indicated by the colour of the marabuzo.
Those candles which exceeded their highs for red marabuzo and lows for green marabuzo before the sufficient continuation length had been reached were counted as failures.
The results revealed that out of a total of 70 marabuzo bars since 5th January 1998, 33 met the criteria for continuation as stipulated in this inital experiment whilst 37 failed.
This seemed to sugest that in practice marabuzo candle’s were not very good at indicating continuation.
Next, to be certain that the results for euro-dollar were not a random occurence I also tested GBP/USD using exactly the same criteria back-tested from the same date.
I found that out of a total of 61 marabuzo candles 31 successfully indicated continuation whilst 30 failed. The results were slightly better, but neither cable nor euro-dollar showed a statistically significant bias towards marabuzo’s signalling continuation.
2nd Experiment – marabuzo with trend
In my 2nd experiment I changed the parameters by introducing MACD as a way of determining the trend, reasoning that it would be better to only test those marabuzo which occurred with the trend.
I only counted as valid green marabuzo’s which occurred when the MACD was above both the zero-line and the signal-line; and red marabuzo’s which occurred when MACD was below the zero-line and below the signal-line as well.
The filter resulted in a sample of 38 marabuzos.
I then applied the tests conditions used to define continuation I had used in the 1st experiment.
The results showed that of the 38 marabuzos which occurred with the trend 21 marked continuation whilst 17 did not.
Using MACD improved the results but not enough: 21 vs 17 was not statistically significant enough to prove a definitive bias.
There was still no proof that this distinguished continuation candle worked.
3rd Experiment – using a wider stop
In the 3rd experiment with marabuzo candles I decided to widen the level which determined a marabuzo had failed. Previously it had been at the highs of the marabuzo in a down-trend and at the lows in an up-trend. I reasoned that perhaps the stop was too tight and I was being unduly harsh in my requirements for the candlestick indicating more or less ‘instant continuation’, so I widened the levels.
According to the new criteria I placed the point at which a marabuzo failed at the length of the marabuzo’s range above the marabuzo high in a down-trend and below the low in an up-trend.
The target for continuation was also expanded and doubled so that it was now twice the range above the high in an up-trend or twice the range below the low in a down-trend. This was necessary to give a roughly 1:1 chance of success or failure.
The new parameters were tested without MACD
The results revealed wider parameters led to an equal 50/50 distribution of 35 winners and 35 losers. Even with a wider ‘stop’ the marabuzo failed to prove a useful tool in indicating continuation.
4th Experiment – wider stop with trend
Finally using both the expanded levels from the 3rd experiment and the MACD as a filter together I tested the data again.
The results showed that 38 candles were defined as marabuzos using MACD and wider success/failure parameters, whilst out of those 23 indicated continuation and 15 did not.
This was an improvment on the the win-rate using MACD alone or wider levels alone but was it statistically significant?
The data revealed that the 23 vs 15 win/loss result was not statistically relevant using scientific standards of statistical significance.
It was, however, close to significance and showed that it was highly unlikely that such a result would be generated randomly. There was only a 12.8% chance that tossing a coin 38 times, for example, would result in 23 heads vs 15 tails (or vis versa). Nevertheless the scientific requirement of 5.0% was not met.
A larger sample would help to decide if there was a bias but it would have to be on a different time-frame or different currency pair, or in the future once EUR/USD had generated more historical data.
Overall there was no way of backing up the assertion that marabuzo candle’s signal continuation in EUR/USD. Even with the aid of MACD determining the trend and filtering out contra-trend examples the data was not statistically significant enough to reveal a bias.