Kruskal-Wallis test

1. Kruskal-Wallis Test

The Kruskal-Wallis test is a non-parametric test used to assess whether several samples originate from the same distribution. It’s a generalization of the Mann-Whitney U test, which is used to compare only two groups. Developed by William Kruskal and W. Allen Wallis in 1952, it’s particularly useful when the assumptions of Analysis of Variance (ANOVA) are not met, specifically concerning the normality of the data. This makes it a powerful tool in fields like technical analysis where financial data often deviates from normal distributions.

When to Use the Kruskal-Wallis Test

The Kruskal-Wallis test is appropriate when:

• You have three or more independent groups (samples).
• The dependent variable is measured on at least an ordinal scale (i.e., data can be ranked). This means the data doesn’t need to be continuous or normally distributed. It can handle discrete data as well, which is common in analyzing trading volumes or price movements.
• The assumptions of ANOVA (normality, homogeneity of variance) are violated. In trading strategies, this is frequently the case due to the presence of outliers (e.g., black swan events) or non-linear price behavior.
• You want to determine if there are statistically significant differences in the medians of the groups. While the test doesn’t directly compare medians, a significant result suggests the groups likely have different medians.

Understanding the Null and Alternative Hypotheses

Like all statistical tests, the Kruskal-Wallis test starts with formulating hypotheses:

• Null Hypothesis (H₀): All samples originate from the same distribution. In simpler terms, there's no significant difference between the groups. For example, if comparing the returns of three different momentum indicators, the null hypothesis would state there's no difference in the median returns of those indicators.
• Alternative Hypothesis (H₁): At least one sample is different from the others. This means at least one group's distribution is stochastically different. If the alternative hypothesis is true, at least one of the moving averages being tested yields significantly different results.

How the Kruskal-Wallis Test Works

The Kruskal-Wallis test is a rank-based test. Here's a step-by-step breakdown of the process:

1. Combine and Rank the Data: All observations from all groups are pooled together and ranked from lowest to highest. Ties are assigned the average rank. This ranking is crucial because it transforms the original data into a scale that can be analyzed without assuming normality. This aligns with the principles of Elliott Wave Theory, where patterns are identified based on relative price movements rather than absolute values.

2. Calculate the Rank Sums: For each group, the ranks of its observations are summed up. This gives you a total rank for each group. For example, if analyzing the performance of three different Fibonacci retracement levels, you'd sum the ranks of the returns generated at each level.

3. Calculate the Test Statistic (H): The test statistic (H) is calculated using the following formula:

   H =  [ (12 / (N * (N + 1))) * Σ(Ri² / ni) ] - 3 * (N + 1)

   Where:

   *   N = Total number of observations across all groups.
   *   Ri = Sum of ranks for the i-th group.
   *   ni = Number of observations in the i-th group.
   *   Σ represents the summation across all groups.

4. Determine the Degrees of Freedom (df): The degrees of freedom for the Kruskal-Wallis test are calculated as:

   df = k - 1

   Where:

   *   k = Number of groups.

5. Determine the p-value: The calculated H statistic is compared to a chi-square distribution with the appropriate degrees of freedom. This yields a p-value. The p-value represents the probability of observing the obtained results (or more extreme results) if the null hypothesis were true. This is similar to how Bollinger Bands represent the probability of price movements relative to a moving average.

6. Make a Decision: If the p-value is less than the chosen significance level (alpha, typically 0.05), the null hypothesis is rejected. This means there is statistically significant evidence to suggest that at least one of the groups is different from the others. If the p-value is greater than alpha, the null hypothesis is failed to be rejected, meaning there isn't enough evidence to conclude a difference between the groups.

Interpreting the Results

A significant Kruskal-Wallis test (p < alpha) indicates that there is a statistically significant difference between the groups. However, it *doesn’t* tell you *which* groups are different from each other. To determine which specific groups differ, you need to perform post-hoc tests.

• Post-Hoc Tests: Common post-hoc tests include the Dunn's test and the Conover-Imbens test. These tests perform pairwise comparisons between groups, adjusting for multiple comparisons to control the family-wise error rate. These are analogous to using multiple support and resistance levels to confirm a trading signal.

• Effect Size: While the Kruskal-Wallis test tells you if there's a significant difference, it doesn’t tell you the *magnitude* of the difference. Effect size measures, such as epsilon-squared (ε²), can be calculated to quantify the practical significance of the findings.

Example: Comparing the Returns of Different Trading Strategies

Let's say a trader wants to compare the performance of three different day trading strategies:

• Strategy A: Based on Relative Strength Index (RSI).
• Strategy B: Based on MACD.
• Strategy C: Based on stochastic oscillator.

The trader collects daily returns for each strategy over a six-month period. The data is not normally distributed, and initial analysis reveals heteroscedasticity (unequal variances). Therefore, the Kruskal-Wallis test is chosen.

1. Data Collection: The daily returns for each strategy are recorded (e.g., Strategy A: 0.5%, -0.2%, 0.8%, Strategy B: 0.1%, -0.5%, 0.3%, Strategy C: 0.2%, 0.0%, 0.6%).

2. Ranking: All 18 returns are combined, ranked from lowest to highest, and ties are assigned average ranks.

3. Rank Sums: The ranks corresponding to each strategy are summed up.

4. H Statistic Calculation: The H statistic is calculated using the formula provided earlier.

5. P-value Determination: The calculated H statistic and degrees of freedom (3-1 = 2) are used to determine the p-value using a chi-square distribution.

6. Decision: If the p-value is less than 0.05, the trader would reject the null hypothesis and conclude that there is a statistically significant difference in the performance of the three strategies. A post-hoc test would then be used to determine which strategies are significantly different from each other.

Advantages and Disadvantages of the Kruskal-Wallis Test

Advantages:

• Non-parametric: Doesn’t require assumptions about the distribution of the data. This is a significant advantage when dealing with financial data, which often isn’t normally distributed.
• Versatile: Can be used with ordinal or continuous data.
• Easy to Understand: The concept of ranking is relatively straightforward.
• Robust: Less sensitive to outliers than parametric tests like ANOVA. This is important in risk management where extreme events can significantly impact results.

Disadvantages:

• Less Powerful: Generally less powerful than parametric tests when the assumptions of parametric tests are met. Meaning, it might fail to detect a true difference when one exists.
• Only Indicates a Difference: Doesn’t identify which specific groups are different, requiring post-hoc tests.
• Sensitive to Ties: A large number of ties can affect the accuracy of the test.
• Focuses on Medians: Primarily tests differences in medians, not means. While medians are often more robust to outliers, means might be more relevant in some situations.

Kruskal-Wallis Test vs. ANOVA

| Feature | Kruskal-Wallis Test | ANOVA | |-----------------|----------------------|------------------------| | Data Type | Ordinal/Continuous | Interval/Ratio | | Distribution | Non-parametric | Parametric (Normal) | | Assumptions | Fewer | More | | Robustness | More | Less | | Power | Lower | Higher (if assumptions met) | | Focus | Medians | Means |

In algorithmic trading, choosing between the Kruskal-Wallis test and ANOVA depends on the characteristics of the data and the specific research question. If the data is clearly non-normal or contains outliers, the Kruskal-Wallis test is the preferred choice.

Software Implementation

The Kruskal-Wallis test is readily available in most statistical software packages:

• R: `kruskal.test()` function.
• Python (SciPy): `scipy.stats.kruskal()` function.
• SPSS: Nonparametric Tests -> Legacy Dialogs -> K Independent Samples.
• Excel: While Excel doesn't have a built-in Kruskal-Wallis test, you can calculate it manually or use add-ins.

Understanding the correct application and interpretation of the Kruskal-Wallis test is crucial for making informed decisions in various fields, especially in areas like market microstructure analysis where data often deviates from standard statistical assumptions. It provides a reliable method for comparing groups when traditional parametric tests are unsuitable. Consider its advantages and disadvantages carefully when selecting the appropriate statistical test for your analysis. Further exploration of candlestick patterns and their statistical significance can complement the findings from a Kruskal-Wallis test. The test's utility extends to evaluating the effectiveness of different chart patterns and their predictive power. Remember to consider volume analysis alongside the Kruskal-Wallis results for a comprehensive understanding of market behavior. Finally, always be mindful of correlation analysis when interpreting results, as relationships between variables can impact the validity of the test.

Time series analysis can be significantly enhanced by applying non-parametric tests like Kruskal-Wallis. Understanding volatility and its impact on trading strategies is crucial, and Kruskal-Wallis can help compare strategies under different volatility regimes. The test is also applicable to evaluating the performance of different portfolio optimization techniques.

Start Trading Now

Sign up at IQ Option (Minimum deposit $10) Open an account at Pocket Option (Minimum deposit $5)

Join Our Community

Subscribe to our Telegram channel @strategybin to receive: ✓ Daily trading signals ✓ Exclusive strategy analysis ✓ Market trend alerts ✓ Educational materials for beginners