Search results

In this bit of code: IIRMA = Fraction * ( 2 * Series - Series[Lag] ) + ( 1 - Fraction ) * IIRMA[1]; The (2 * Series - Series[Lag]) adds a momentum component to the output, which reduces lag, but also increases noise too. Try using this bit of code: IIRMA = Fraction * ( 0.5 * (Series+Series[1]...

I have attached the technical article. You will note the author recommends using FIR filters, but there is no harm in extending the method to IIR filters.

Yes. You can certainly use more poles in your filter design, but in reverse filtering situations, you will need to estimate more days into the future with higher order IIRs, which introduces more chances for errors in the output.

No, I just want to help folks. For my next tip, to get zero lag or even negative lag, research forward-backward filtering using emas (ie one-pole infinite impulse response filters.) You will get outstanding filtering performance, which can be improved even further by sharpening your reverse ema.

Here is a simple trick from the world of digital signals processing to improve the filtering performance of your MAs. 1. Filter the input signal, x(n) i.e. price, once with your moving average 2. Double the moving average value to obtain w(n) 3. Subtract w(n) from 3*x(n) to obtain u(n) 4...

Kalman Filter Tutorial Some say there is no noise in finance, and that it all signal. However, it is just a model.

Measuring the market dominant cycle and then using that data to adjust or tune indicators is what John Ehlers was all about for the past 20 plus years. However, if you read his last interview with S&C, he basically dismisses cycle measurement. He believes cycles are too ephemeral in price data...

Which is why you should read "Whiter is Brighter." It explains why a 2 day ROC is preferred over a 1 day.

The same physics applies. The higher the frequency cutoff of the filter, the less group delay. The less group delay a filter has, the less attenuation ability. A lot of people over the years have written about Kalman filters in finance. A Kalman filter is really an estimator rather than a...

I use a cyclical indicator "Whiter is Brighter". I have yet to find an MA I am happy (and profitable) with. For those who do offline analysis, I think the ultimate MA is an IIR in forward-backward mode, ala Gustafsson, a true zero group delay filter.

There is a reason moving averages work sometimes and other times not. MAs are in reality lowpass digital filters. The two basic types are finite impulse response (FIR) and infinite impulse response (IIR). FIR filters are your typical MA that everyone knows, and IIR filter are what people call...

Try this article on for on for size.

To send you down the rabbit hole on your journey to learn about moving averages, read "Rocket Science for Traders", by John Ehlers. The TL,DR version is that moving averages are lowpass digital filters. The two fundamental types are finite impulse response (FIR) and infinite impulse response...

Yes, but in financial markets you get paid to take risks. The difference between reality and the results from your model are where the risks lie. Markets will never be modeled perfectly because the complexity is too great, just as in weather. What happens when central bank models of the economy...

Trying to find cause and effect for every human decision that moves the market is a fruitless exercise and surely a path to madness. I like the simple pink noise model of price, k/pow(f,alpha), where you set k and alpha equal to 1 for simplicity sake. This gives a power spectral density of 6dB...

After many years of trying to work with trend following and moving averages of different flavors, I have moved on (profitably) to cycles. I personally like Ehlers' pink noise model of the market.

For your ema, try this formula, and play around with values for alpha and gain, typical values might be alpha = 0.2 and gain = 0.5: y[0] = alpha*(x[0]+gain*(x[0]-y[1]))+(1-alpha)*y[1] Also, read the article by John Ehlers and Ric Way, "Zero Lag" to find an algorithm to optimize gain.

These are the reasons a person needs a TA method with enough mathematical exactness that it can be programmed into an algorithm. The algorithm can then be back tested to provide some assurance of positive expectancy once the method were to be used live. The algorithm could be as simple as using...

Sorry I'm responding to this thread late, but earth_imperator is right, this model assumes price is pink noise and not GBM. That is ok, as I prefer the pink noise model for the TA I do.

In my experience, no they don't. In fact they all stink, in backtests and realtime. What one needs is an indicator that combines zero group delay (look it up) with acceptable attenuation ability. I may have discovered one such, but that remains to be proven.