Using the MACD histogram or the MACD fast and slow lines give you the same information, really. Regardless how you look at it, it highlights MA crossovers, divergences, momentum highs/lows, etc. It is a matter of preference. I used to use MACD as a histogram, because that is the way I first learned to use it, from Alexander Elder's books. I removed it because I wanted to simplify my charts. Now, increasing momentum, as highlighted in the MACD, does provide a leading signal and usually leads to new price highs/lows. This market tendency alone is an extremely high probability trade that works in all markets and timeframes. Test it out for yourself and see if it is not true.
That's great Buy1Sell2, but not because of the MACD histogram. You're either using intuition, something else, or anticipating what MACD will do, which could lead to whipsaws.
I don't care to argue my point, but I believe you misunderstood it. So ... 1. You can use similar mathematical formulae to accomplish various things. I disagree with stockcharts.com's description of MACD, and I believe that author doesn't know what he's writing about. MACD is a band pass filter with an added smoothing and delay. Using it incorrectly will reduce its efficiency, and people will conclude that MACD doesn't work. 2. When I wrote about optimizing parameters, I didn't mean curve fitting. I meant that price action in various time frames has different characteristics, and that it's better to optimize each indicator's parameters accordingly.
I would argue that the base macd is the difference between two bandpass filters (two moving averages). Stockcharts is correct that you subtract one from another. SC is also correct that this provides a measure of momentum. If you doubt it take the case where instead of 12/26 as the base parameter you use 1/26. In this case it behaves like a CCI26 (ignoring any issues of simple vs exponential moving averages and close vs average prices) which is simply the momentum of last price away from a moving average of price. This is not to dispute your "waves approach" just to point out that SC's interpretation is valid for them. I agree with cnms in that the key for something like macd is to find the right 3 ma measures to represent the wave/swing behaviour in the contract you trade and 12/26/9 may not be it any more than I would use the same base ema for currencies and stock indexes. Variation doesnt imply overfitting, just fitting; overfitting is avoided by good "system" design and testing.
kiwi_trader, I think you meant MAs are low-band pass filters (filter out higher frequencies). This makes the MACD a band pass filter. Using higher values for EMAs (dropping to a lower time frame for the same price action wave), brings them closer to SMAs, hence better filters.
The irony here is that if you are looking at historic prices (and/or volume and/or breadth and/or open interest, where applicable) in the calculation of your "statistical edges," then you are resorting to TA. That is what TA is. It need not necessarily include either tired, boilerplate indicators or subjective chart patterns. The use of historical data of price et al implies the use of TA. Unless your statistical analyses include fundamental data (econometrics), then you are simply resorting to a variation of TA. Unless, of course, you are looking to redefine the core definition of TA.
I have always thought that as well--that people who say I don't use TA, I just look at price are really practicing TA without wanting to say so. I would add one note: tired boilerplate indicators still work when used properly. I have used them for years and they are still valid today.