The D0 collaboration at FermiLab has recently measured the top-quark pair forward-backward asymmetry in $\bar p p \to t \bar{t} X$ reactions as a function of the $t\bar{t} $ invariant mass $M_{t\bar{t}}$. The D0 result for $A_{\rm FB}(M_{t\bar{t}}>650\, {\rm GeV})$ is smaller than $A_{\rm FB}(M_{t\bar{t}})$ obtained for small values of $M_{t\bar{t}}$, which may indicate an "increasing-decreasing" behavior for $A_{\rm FB}(M_{t\bar{t}}>M_{\rm cut})$. This behavior is not explained using conventional renormalization scale-setting, even by a next-to-next-to-leading order (N$^2$LO) QCD calculation -- one predicts a monotonically increasing behavior. In the conventional scale-setting method, one simply guesses a single renormalization scale $\mu_r$ for the argument of the QCD running coupling and then varies it over an arbitrary range. However, the conventional method has inherent difficulties. ...... In contrast, if one fixes the scale using the Principle of Maximum Conformality (PMC), the resulting pQCD predictions are renormalization-scheme independent since all of the scheme-dependent $\{\beta_i\}$-terms in the QCD perturbative series are resummed into the QCD running couplings at each order. ...... In this paper we show that if one applies the PMC to determine the top versus anti-top quark forward-backward asymmetry by properly using the pQCD predictions up to N$^2$LO level, one obtains the predictions without renormalization scheme or scale ambiguities. ...... In addition, the PMC prediction for $A_{\rm FB}(M_{t\bar{t}}> M_{\rm cut})$ shows an "increasing-decreasing" behavior for increasing values of $M_{\rm cut}$ which is not observed in the NLO and N$^2$LO predictions for $A_{\rm FB}(M_{t\bar{t}}> M_{\rm cut})$ with conventional scale-setting. This behavior could be tested by the future more precise measurements at the LHC.