Just check the angle of attack with azimuthal angle of VAWT (actually, lower TSR implies very high angle of attack), and you can classify the contribution of lift and drag to the rotational torque with the azimuthal angle. Finally, you'll find the torque can be negative depending on the characteristics of wing of lift and drag in the very high angle of attack. Probably, it may be related with the stall characteristics.
Can you look at the pressure contours near the blades? And the velocity vectors? You should be able to estimate the angle of attack from the vectors and see if the pressure contours make sense according to experiments on airfoils (highly documented)
1. If you are using steady state simulation (MRF) you may end with negative torque, which is basically is the value of starting torque. Which is almost very low value (cp = 0.04 or so), but it is there due to inertia of turbine. This is common issue of HAWT that you need to provide starting torque to start its operation.
2. At very value of TSR, it is also possible that turbine is not able to produce enough torque values to sustain it motion, therefore it is again giving you starting torque values which are definitely negative.
3. It is not necessary if others are not getting -ve values, you must not. It may be cambered airfoil etc. Or may these are designed for lower values of TSR depending on solidity.
4. Sometimes, if you setup every thing correctly, you will get -ve torque values in initial stages of unsteady simulation, but once it is settled you will start to get positive mean torque values.
Note: This question was asked few years back, so it will not be helpful to original poster, but it will helpful to readers who come here for help on same topic.