@hernavs I did my bachelor's in time series prediction using a (rather exotic) RNN. I'm not up to date with current deep neural net developments though.
As for your question, I think this is plausible in some cases. Since trained CNN's are in principle time-invariant filters which are used in time series processing and pattern detection for decades. And as we know, neural nets are universal function approximators so in theory, a CNN may learn any time-invariant filter. So, as long as stationarity is not violated (i.e. your pattern doesn't change over time), CNN's might be just as good as RNN's, at least in theory.
LSTM's have inherent biases decrease conversion time in learning time series. Maybe those biases hold LSTM's back in some cases, and RNN's can shine in such cases?