Forecasting stock market trends in often challenging due to the randomness and volatility of the nature of its movement owing to several factors. Often it is difficult to formulate a numerical counterpart for the diverse set of governing variables which can potentially impact stock movement. Moreover, collecting such data through indirect data mining techniques may often contain substantial biased and skewed opinions pertaining to a particular sample set. Thus, a naturally pragmatic approach seems to be designing a model trained on historical data, which is able to forecast both in and out sample. This would render high degree of robustness and practical utility to the developed model. This paper presents an approach to combine multi-resolution analysis with deep neural networks to forecast the future movement of stock markets. The multi-resolution analysis is used to remove effects of baseline noise inherent to time series stock datasets, followed by pattern recognition using deep neural networks. A diverse set of S & P datasets have been chosen for analysis. The forecasting accuracy, error rates and regression have been used compare the performance of the model against benchmark existing models.