Under a random walk, the variance of two-day returns should be twice the variance of one-day returns, of five-day returns five times, and so on. The variance ratio is the observed long-horizon variance divided by that prediction.
A ratio above 1 indicates positive autocorrelation, returns building on each other, which is trending behaviour. Below 1 indicates reversal. A ratio of 0.88 at the ten-day horizon, for instance, suggests modest mean reversion over two weeks.
It shares the weaknesses of every test here: sensitive to the horizon chosen, low power on short samples, and vulnerable to being run at many horizons until one looks significant. Pick the horizon from the trade design, not from the results.
Related: hurst-exponent, autocorrelation, random-walk, p-value