F0 downtrend observed in spontaneous monologues in the Corpus of Spontaneous Japanese was analyzed with special attention to the modeling of final lowering. In addition to the previous finding that the domain of final lowering covers all tones in the final accentual phrase, it turned out that the last L tone in the penultimate accentual phrase played important role in the control of final lowering. It is this tone that first reached the bottom of the speaker's pitch range in the time course of utterance; it also turned out that the phonetic realization of this tone is the most stable of all tones in terms of the F0 variability. Regression model of F0 downtrends is generated by generalized linear mixed-effect modeling and evaluated by cross-validation. The mean prediction error of z-normalized F0 values in the best model was 0.25 standard deviation.