Using Mann-Kendall Test on Litecoin and More

There are many cryptocurrencies out there, but I have been mostly playing with Litecoin. I recently asked myself whether there would be a statistical model that I could use to show trends.

To make things easier, I assumed there are no variables that can significantly affect the prices. However, we know that it's not true. Prices are influenced considerably by the macroenvironment and speculations, among other things. I'll have to revisit those variables later after creating a simpler analysis.

I had to think about what models could be used in other cases that I could apply here. After searching on Google, I realized the Mann-Kendall test can be used on cryptocurrencies to identify the trend. It tests the data to see how the prices move up or down consistently. I used a Python package to run the analysis and collected the data for several weeks using Selenium to scrape from a website. Moreover, the model could identify trends in complaints, attrition, and more for business use cases.

I performed the analysis for different durations. For instance, a more extended period would show that the trend is decreasing, but the trend is increasing in a shorter period. It shows the relative significance of the variance of y and x. For instance, the Litecoin price could have a significantly higher variance in a month than a week. It can also show that the price is trending down in the long run (relatively speaking); therefore, it tells us that there is a higher risk to investing. 

As you can see from the graph and outputs, the middle slope starts from the intercept to the current price. The actual price (in red) is moving consistently within the slope. This is based on approximately 10,000 minutes or 166+ hours or almost 7 days. 

I added buffer slope lines by 2% differences to give me an idea of what price I should invest next. If the trend holds, then I could invest around $110 per coin. I would have to add more analysis to consider the duration. The duration could change the buy price at any time. An analysis done on a longer duration can also show that the trend decreases. 

To make this model work well with my analysis,  I would need to add multiple duration scenarios to compare the trend and the statistical significance, such as the z-score. Furthermore, I can add different variables from news outlets to create natural language classifiers. Those categories could be used to understand correlations and potentially predict prices. 

I can definitely see how the Mann-Kendall test can be used to solve business problems. For instance, complaints at a call center could potentially be predicted. HR could also use this model to understand what may cause higher attritions.

#analysis #analytics #dataengineering #stablecoin #litecoin #bitcoin #crypto