Time series analysis has been the subject of extensive interest in many fields ofstudy ranging from weather forecasting to economic predictions, over the past twocenturies. It has been fundamental to our understanding of previous patterns withindata and has also been used to make predictions in both the short and long termhorizons. When approaching such problems researchers would typically analyzethe given series for a number of distinct characteristics and select the most ap-propriate technique. However, the complexity of aligning a set of characteristicswith a method has increased in complexity with the advent of Machine Learningand the introduction of Multi-Step Ahead Prediction (MSAP). We examine themodel/strategy approaches which are currently applied to conduct multi-step aheadprediction in time series data and propose an alternative MSAP strategy known asMulti-Resolution Forecast Aggregation.Typically, when researchers propose an alternative strategy or method, they demon-strate it on a relatively small set of time series, thus the general breath of use isunknown. We propose a process that generates a diverse set of synthetic time se-ries, that will enable a robust examination of MRFA and other methods/strategies.This dataset in conjunction with a range of popular prediction methods and MSAPstrategies is then used to develop a meta learner that estimates the normalized meansquare error of the prediction approach for the given time series