This thesis proposes a new stochastic model for the daily number of transactions with stocks in the Brazilian capital market. The model, which we call ZINB-HF (Zero – Inflated Negative Binomial Harvey- Fernandes), is an extension of a local scale model for count data. In it, the daily number of transactions follows a serially correlated negative binomial distribution. The negative binomial is generated through three independent microstructural models of the transaction process. The main model is based on the concept of random ruptures in the returns generating process, which occur in an alternative time frame of reference, defined by the transactions themselves. The second microstrutural model suggested, to justify the usage of a negative binomial points to the concept of non observed heterogeneity in the frequency of trading by market agents. The third one is based on the concept of positive contagion related to a self-reinforcing process of liquidity on the stock. The serially correlated negative binomial of the local scale model is supposed zero inflated, which means that the probability of zero transactions in any day is larger than indicated by a pure negative binomial, due to the existence of forced breaks in trading. The proposed ZINB-HF model is used as a directing process of daily returns of the stock, in the framework of a time deformation model. The work concludes with a case study a representative Brazilian stock.