Quote from dottom:
If you are daytrading you will want an MA that can generate a signal much faster than what SMA or EMA will give you. You can look at the some of the adaptive MA's but eventually you will settle on something based on DSP. You not only want to be able to filter out more noise, but also maintain less lag. The noise-lag relationship is usually a sacrifice between one and the other (smoothness vs. lag by expanding sample window). But various DSP techniques allow you to optimize both making fewer sacrifices.
The best publicly available MA is Jurik's JMA.
Regarding daytrading, the reason an SMA or EMA may not give you a valid signal is in the beginning of the trading session your MA will be out of sync. Even if there is no gap, the dynamics of each morning session are different from the previous day's ending session, yet an SMA or EMA will simply treat the pervious day's price data as if it were today's. You need something that can adjust quickly and filter out the noise. This is especially true if there is a large gap.
If you are building dynamic non-linear trading models and using an MA as a proxy for price because bar-to-bar price changes are too noisy, then you want to use the best possible MA you can find.
A better MA does make a difference, but it depends on your method, timeframe, and how you use it. Eg. if you must two MA's as just a simple trend filter, then it doesn't matter as much.