I tested it and can confirm the findings from @cajetan.rodrigues. However, this does not seem to be _not_ documented. The documentation clearly states that using a positive value for radius results in a counter-clockwise direction, whereas a negative number results in a clockwise direction. This is true.

Now, let's have a look at using radius _together_ with extent, which is not (clearly) documented as this point. Using a positive value results in the same direction, but in a not full drawn circle. This seems intuitive and is documented:

    If extent is not given, draw the entire circle.
    If extent is not a full circle, one endpoint of the arc is the current pen position.

Using a negative value changes the direction of radius, which does also seem reasonable and intuitive.

My suggestion is not to replace "radius" by "extent" as suggested by @guchao as it is IMHO wrong. But I agree that it would be good to leave a note there or to add examples demonstrating the behaviour of different (signed) values of radius and extent, so the intent and behaviour is clearly documented.