This is just fitting on noise. The vast majority of movements are random and no more predictable than a coin flip. Before training, your job is to extract that

This is just fitting on noise. The vast majority of movements are random and no more predictable than a coin flip. Before training, your job is to extract that extremely weak signal, then train. Try generating a time series in Excel with Brownian noise, watch as it is indistinguishable from price charts.