Doji Significance?

I. Introduction & Overview

II. Results

III. Conclusion

*UPDATE 11/25/2021: Updated test conditions, results, and conclusions to be more in-line with proper statistical assumptions, more accurate, and more comprehensible than before

I. Introduction & Overview

In traditional trading education, we are taught that a candlestick that opens and closes at roughly the same price has significance. This candlestick is known as a "doji," named by 18th century Japanese rice traders. Do doji candlesticks really have some sort of predictive power?

The initial hypothesis behind the doji is that because price opened and closed at virtually the same price (or at least fairly close), it's a symbol of indecision and can result in a reversal from its current trend.

This study will be performed on the SPDR S&P 500 ETF Trust (SPY) since 2003 on a daily timeframe. We will analyze the next-day's close following a doji to determine if there is any statistical significance against any other type of candlestick. In order to add some trend context to the study, we will also tests the results depending on the direction it closed in relation to the last one and two days. For 1 higher/lower close, the previous close will be controlled to ensure that their results are not mixed with the 2 higher/lower close test condition. Below is an example of the comparisons that will be made:

First, we need to quantify what is considered a doji. Two factors will be taken into account: position and width.

To measure position, a metric called Interbar Strength (IBS) will be used. To measure width, we will apply the following formula below:

IBS measures the position of the candle, 0 meaning the closing price is equal to its daily low and 1 meaning the closing price is equal to its high. The width formula is a custom formula to measure the width of the open and closing prices. In order for a candle to be considered a doji, its IBS must be in the range of [0.25, 0.75] and its width must be less than or equal to 0.10

Take a look at some examples below of what we will and will not be considered a doji:

Once all doji tests have been run, they will be compared against any other type of candle (other than a doji) in that same context to determine if there is any statistical significance via t-test against means of percent change. All returns distributions have been verified to be mostly normal and that the ratio of variances between each group and its appropriate benchmark are less than 4 to assume equal variance in significance tests.

II. Results

When running the test and reporting means as percentages, the results are found below. The image containing the test types will also be posted again for reference.

From the output table, we can see that there is no statistical significance in a doji predicting the next-day's price move, nor predicting a reversal from it's previous one or two-day trend. There is also no practical significance as the largest difference in means was -0.19%. The expectation to receive a 19bps difference in returns is unlikely to add enough value to a strategy in sacrifice to the addition of an extra parameter and therefore increasing model complexity.

III. Conclusion

Unlike what traditional trading education teaches - we conclude that we fail to find any statistical or practical significance in the doji candlestick's ability to predict direction or a change in direction compared to any other candlestick type.