Can the -2LogLikelihood value be negative?

Yes.

At ITW 2014, Michael Neale explained why -2LogLikelihood value be negative.

Typically the likelihood density distribution has a mean of around .5 whose log value is negative. That's why it's multiplied by -2 to make the model fit measure a positive value. However, when the likelihood value is more than 1, its log value would be positive, which makes -2LogLikelihood a negative value when it's multiplied by -2. This happens when the variance distribution is narrow, which creates a high value of likelihood.

Sarah Medland offered a pretty simple solution: Multiply the dependent values by 100 to wide the range of raw data distribution which increases the range of variances and often turn -2LogLikelihood into a positive value consequently.

It's perfectly OK to have a negative value. Most importantly, what matters is the comparison of two model fit indices as opposed to the evaluation of absolute fit measures.