Today, I want to talk about my small study on the relationship between an indicator called the Relative Strength Index (RSI) and the future price movement.

It is commonly believed that low RSI readings can indicate oversold state, while high readings can indicate overbought state. In fact, many examples can be found where, after the RSI indicator reached low levels of around 30 or 20 and below, the price began to rise. Conversely, after reaching high levels, the price tended to decline.
Here is an example that illustrates this apparent pattern:

Visually, everything seems to be correct, and this gives many traders reason to consider the RSI indicator as one of the significant arguments for making a trading decision.

However, let's not rely on what we see and what seems obvious. After all, we can conduct some research and obtain facts.
As with any other analysis, we need to start with data collection, parapapapa i'm lovin' it))).

Today, data collection does not pose significant difficulties. For example, using the Binance API, we can obtain almost any historical data.
I decided to analyze historical values of the minute timeframe. I chose the minute timeframe because, in retrospect, it smooths out strong fluctuations. What looks like a one-hour timeframe gap with a single red candle will be smoother on the minute timeframe. Moreover, the general conclusions of the analysis can be extrapolated to other timeframes.

I obtained data on 5000 candles, which corresponds to approximately a three-and-a-half-day period, for the XRPBUSD trading pair. Of all the data, I am interested in high, low, and closing prices.

Fragment of received data

Getting prices is simple, but Binance does not provide RSI data, so I had to calculate this indicator myself.

To do this, I used a method that involves calculating the Exponential Moving Average (EMA), then RS, and finally RSI for each candlestick. There are several approaches to calculating RSI, and I am not sure which method Binance uses. Additionally, I do not know which smoothing coefficients they use, so I needed to compare my results directly with Binance data.

As expected, there were some discrepancies, but for the most part they were within a few units or less, and could not have a significant impact on further analysis.

Now that we have historical data on prices and RSI value, the most interesting part begins!

For each price and RSI value, we can not only assume the direction of future price movements, we can "look into the future" and know exactly where the price will go.
The next step is to establish criteria for what can be considered an upward or downward movement. I have decided to consider a price increase of 0.5% or more as an upward movement, and accordingly consider a price decrease of 0.5% or more as a downward movement. This will depend on which occurs first.
That is, for to each piece of information on the current price and RSI, further price fluctuations will result in a price change of 0.5% or more, up or down. Depending on which occurred first, I evaluated the movement. If the price increased in the future, I marked it with a value of 1, and if the price decreased, I marked it with a value of -1.

Now we have the following information: Price, RSI, and the future direction of the price: 1 if the price increased in the future and -1 if the price decreased in the future.

We can now proceed directly to the analysis.
I decided to start with calculating Pearson correlation coefficients, which means establishing whether there is a correlation between the current RSI value and the future direction of the price.

Despite expecting something similar (which is why I started this research), the result surprised me. The Pearson correlation coefficient was only -0.0265, indicating that there is no clear relationship between RSI and price movement. I would like to remind you that a coefficient value of at least 0.5 is required to speak of any correlation, and a strong correlation can be said to exist when this coefficient approaches 1. Result I got, shows almost complete lack of correlation, value is at margin of statistical error.

This made me doubt the representativeness of the sample, so I redid everything for 12,000 candles. As a result, I obtained a correlation coefficient of -0.023, which only confirmed the first result.

To visualize the level of dependence (or almost complete absence thereof), I performed a cluster analysis. I assigned each RSI value to one of the classes (from 0 to 9.99, from 10 to 19.99, from 20 to 29.99, and so on). Then I calculated the number of 1 and -1 results for each class (i.e. the number of cases where the price rose in the future and the number of cases where the price fell in the future). I also calculated the ratio of the number of growth cases to the total number of cases in the class. The results were presented in a table format.

XRPBUSD cluster analysis results

As we can see, when the RSI values range from 0 to 19.99 or from 40 to 69.99, the likelihood of price increase or decrease is practically the same. Some sort of pattern, if we can call it that, emerges in the range of 20 to 39.99, but it is too insignificant to rely on.
Only for the RSI range of 70 to 89.99 can we talk about some dependence. We see that after reaching such values, in most cases, the price decreases. However, this is far from the majority, and therefore even at such values, the RSI indicator can be considered only as one of the factors for decision-making. And by no means a decisive factor.

To be more objective, I conducted the same analysis for some other trading pairs. Perhaps this is just a feature that is peculiar only to XRPBUSD? The result is before you. As we can see, a similar situation occurs with other assets.

BTCBUSD cluster analysis results

ETHBUSD cluster analysis results

BNBBUSD cluster analysis results

I think many people have been in situations where the RSI indicator looked good, and it seemed to indicate a high probability of further price movement in the desired direction. And after entering into a deal, they were surprised why the price did not go there, attributing it to a coincidence or some fundamental events they did not know about. But now you know why this is happening. And I hope this will encourage you to make more responsible decisions.

Thank you for your attention, I would greatly appreciate it if you could share your thoughts on the results of my research. Perhaps you have ideas on the dependencies of other indicators, conditions, or their combinations that can be further researched.