Indicator
Nadaraya-Watson: Rational Quadratic Kernel (Non-Repainting)What is Nadaraya–Watson Regression?
Nadaraya–Watson Regression is a type of Kernel Regression, which is a non-parametric method for estimating the curve of best fit for a dataset. Unlike Linear Regression or Polynomial Regression, Kernel Regression does not assume any underlying distribution of the data. For estimation, it uses a kernel function, which is a weighting function that assigns a weight to each data point based on how close it is to the current point. The computed weights are then used to calculate the weighted average of the data points.
How is this different from using a Moving Average?
A Simple Moving Average is actually a special type of Kernel Regression that uses a Uniform (Retangular) Kernel function. This means that all data points in the specified lookback window are weighted equally. In contrast, the Rational Quadratic Kernel function used in this indicator assigns a higher weight to data points that are closer to the current point. This means that the indicator will react more quickly to changes in the data.
Why use the Rational Quadratic Kernel over the Gaussian Kernel?
The Gaussian Kernel is one of the most commonly used Kernel functions and is used extensively in many Machine Learning algorithms due to its general applicability across a wide variety of datasets. The Rational Quadratic Kernel can be thought of as a Gaussian Kernel on steroids; it is equivalent to adding together many Gaussian Kernels of differing length scales. This allows the user even more freedom to tune the indicator to their specific needs.
The formula for the Rational Quadratic function is:
K(x, x') = (1 + ||x - x'||^2 / (2 * alpha * h^2))^(-alpha)
where x and x' data are points, alpha is a hyperparameter that controls the smoothness (i.e. overall "wiggle") of the curve, and h is the band length of the kernel.
Does this Indicator Repaint?
No, this indicator has been intentionally designed to NOT repaint. This means that once a bar has closed, the indicator will never change the values in its plot. This is useful for backtesting and for trading strategies that require a non-repainting indicator.
Settings:
Bandwidth. This is the number of bars that the indicator will use as a lookback window.
Relative Weighting Parameter. The alpha parameter for the Rational Quadratic Kernel function. This is a hyperparameter that controls the smoothness of the curve. A lower value of alpha will result in a smoother, more stretched-out curve, while a lower value will result in a more wiggly curve with a tighter fit to the data. As this parameter approaches 0, the longer time frames will exert more influence on the estimation, and as it approaches infinity, the curve will become identical to the one produced by the Gaussian Kernel.
Color Smoothing. Toggles the mechanism for coloring the estimation plot between rate of change and cross over modes.
Indicator
Possible RSI [Loxx]Possible RSI is a normalized, variety second-pass normalized, Variety RSI with Dynamic Zones and optionl High-Pass IIR digital filtering of source price input. This indicator includes 7 types of RSI.
High-Pass Fitler (optional)
The Ehlers Highpass Filter is a technical analysis tool developed by John F. Ehlers. Based on aerospace analog filters, this filter aims at reducing noise from price data. Ehlers Highpass Filter eliminates wave components with periods longer than a certain value. This reduces lag and makes the oscialltor zero mean. This turns the RSI output into something more similar to Stochasitc RSI where it repsonds to price very quickly.
First Normalization Pass
RSI (Relative Strength Index) is already normalized. Hence, making a normalized RSI seems like a nonsense... if it was not for the "flattening" property of RSI. RSI tends to be flatter and flatter as we increase the calculating period--to the extent that it becomes unusable for levels trading if we increase calculating periods anywhere over the broadly recommended period 8 for RSI. In order to make that (calculating period) have less impact to significant levels usage of RSI trading style in this version a sort of a "raw stochastic" (min/max) normalization is applied.
Second-Pass Variety Normalization Pass
There are three options to choose from:
1. Gaussian (Fisher Transform), this is the default: The Fisher Transform is a function created by John F. Ehlers that converts prices into a Gaussian normal distribution. The normaliztion helps highlights when prices have moved to an extreme, based on recent prices. This may help in spotting turning points in the price of an asset. It also helps show the trend and isolate the price waves within a trend.
2. Softmax: The softmax function, also known as softargmax: or normalized exponential function, converts a vector of K real numbers into a probability distribution of K possible outcomes. It is a generalization of the logistic function to multiple dimensions, and used in multinomial logistic regression. The softmax function is often used as the last activation function of a neural network to normalize the output of a network to a probability distribution over predicted output classes, based on Luce's choice axiom.
3. Regular Normalization (devaitions about the mean): Converts a vector of K real numbers into a probability distribution of K possible outcomes without using log sigmoidal transformation as is done with Softmax. This is basically Softmax without the last step.
Dynamic Zones
As explained in "Stocks & Commodities V15:7 (306-310): Dynamic Zones by Leo Zamansky, Ph .D., and David Stendahl"
Most indicators use a fixed zone for buy and sell signals. Here’ s a concept based on zones that are responsive to past levels of the indicator.
One approach to active investing employs the use of oscillators to exploit tradable market trends. This investing style follows a very simple form of logic: Enter the market only when an oscillator has moved far above or below traditional trading lev- els. However, these oscillator- driven systems lack the ability to evolve with the market because they use fixed buy and sell zones. Traders typically use one set of buy and sell zones for a bull market and substantially different zones for a bear market. And therein lies the problem.
Once traders begin introducing their market opinions into trading equations, by changing the zones, they negate the system’s mechanical nature. The objective is to have a system automatically define its own buy and sell zones and thereby profitably trade in any market — bull or bear. Dynamic zones offer a solution to the problem of fixed buy and sell zones for any oscillator-driven system.
An indicator’s extreme levels can be quantified using statistical methods. These extreme levels are calculated for a certain period and serve as the buy and sell zones for a trading system. The repetition of this statistical process for every value of the indicator creates values that become the dynamic zones. The zones are calculated in such a way that the probability of the indicator value rising above, or falling below, the dynamic zones is equal to a given probability input set by the trader.
To better understand dynamic zones, let's first describe them mathematically and then explain their use. The dynamic zones definition:
Find V such that:
For dynamic zone buy: P{X <= V}=P1
For dynamic zone sell: P{X >= V}=P2
where P1 and P2 are the probabilities set by the trader, X is the value of the indicator for the selected period and V represents the value of the dynamic zone.
The probability input P1 and P2 can be adjusted by the trader to encompass as much or as little data as the trader would like. The smaller the probability, the fewer data values above and below the dynamic zones. This translates into a wider range between the buy and sell zones. If a 10% probability is used for P1 and P2, only those data values that make up the top 10% and bottom 10% for an indicator are used in the construction of the zones. Of the values, 80% will fall between the two extreme levels. Because dynamic zone levels are penetrated so infrequently, when this happens, traders know that the market has truly moved into overbought or oversold territory.
Calculating the Dynamic Zones
The algorithm for the dynamic zones is a series of steps. First, decide the value of the lookback period t. Next, decide the value of the probability Pbuy for buy zone and value of the probability Psell for the sell zone.
For i=1, to the last lookback period, build the distribution f(x) of the price during the lookback period i. Then find the value Vi1 such that the probability of the price less than or equal to Vi1 during the lookback period i is equal to Pbuy. Find the value Vi2 such that the probability of the price greater or equal to Vi2 during the lookback period i is equal to Psell. The sequence of Vi1 for all periods gives the buy zone. The sequence of Vi2 for all periods gives the sell zone.
In the algorithm description, we have: Build the distribution f(x) of the price during the lookback period i. The distribution here is empirical namely, how many times a given value of x appeared during the lookback period. The problem is to find such x that the probability of a price being greater or equal to x will be equal to a probability selected by the user. Probability is the area under the distribution curve. The task is to find such value of x that the area under the distribution curve to the right of x will be equal to the probability selected by the user. That x is the dynamic zone.
7 Types of RSI
See here to understand which RSI types are included:
Included:
Bar coloring
4 signal types
Alerts
Loxx's Expanded Source Types
Loxx's Variety RSI
Loxx's Dynamic Zones
Indicator
Gaussian Filter MACD [Loxx]Gaussian Filter MACD is a MACD that uses an 1-4 Pole Ehlers Gaussian Filter for its calculations. Compare this with Ehlers Fisher Transform.
What is Ehlers Gaussian filter?
This filter can be used for smoothing. It rejects high frequencies (fast movements) better than an EMA and has lower lag. published by John F. Ehlers in "Rocket Science For Traders". First implemented in Wealth-Lab by Dr René Koch.
A Gaussian filter is one whose transfer response is described by the familiar Gaussian bell-shaped curve. In the case of low-pass filters, only the upper half of the curve describes the filter. The use of gaussian filters is a move toward achieving the dual goal of reducing lag and reducing the lag of high-frequency components relative to the lag of lower-frequency components.
A gaussian filter with...
one pole is equivalent to an EMA filter.
two poles is equivalent to EMA ( EMA ())
three poles is equivalent to EMA ( EMA ( EMA ()))
and so on...
For an equivalent number of poles the lag of a Gaussian is about half the lag of a Butterworth filters: Lag = N * P / (2 * ¶2), where,
N is the number of poles, and
P is the critical period
Special initialization of filter stages ensures proper working in scans with as few bars as possible.
From Ehlers Book: "The first objective of using smoothers is to eliminate or reduce the undesired high-frequency components in the eprice data. Therefore these smoothers are called low-pass filters, and they all work by some form of averaging. Butterworth low-pass filtters can do this job, but nothing comes for free. A higher degree of filtering is necessarily accompanied by a larger amount of lag. We have come to see that is a fact of life."
References John F. Ehlers: "Rocket Science For Traders, Digital Signal Processing Applications", Chapter 15: "Infinite Impulse Response Filters"
Included
Loxx's Expanded Source Types
Signals, zero or signal crossing, signal crossing is very noisy
Alerts
Bar coloring
Indicator
STD-Filtered, Gaussian Moving Average (GMA) [Loxx]STD-Filtered, Gaussian Moving Average (GMA) is a 1-4 pole Ehlers Gaussian Filter with standard deviation filtering. This indicator should perform similar to Ehlers Fisher Transform.
The purpose of the standard deviation filter is to filter out noise by and by default it will filter 1 standard deviation. Adjust this number and the filter selections (price, both, GMA, none) to reduce the signal noise.
What is Ehlers Gaussian filter?
This filter can be used for smoothing. It rejects high frequencies (fast movements) better than an EMA and has lower lag. published by John F. Ehlers in "Rocket Science For Traders". First implemented in Wealth-Lab by Dr René Koch.
A Gaussian filter is one whose transfer response is described by the familiar Gaussian bell-shaped curve. In the case of low-pass filters, only the upper half of the curve describes the filter. The use of gaussian filters is a move toward achieving the dual goal of reducing lag and reducing the lag of high-frequency components relative to the lag of lower-frequency components.
A gaussian filter with...
one pole is equivalent to an EMA filter.
two poles is equivalent to EMA(EMA())
three poles is equivalent to EMA(EMA(EMA()))
and so on...
For an equivalent number of poles the lag of a Gaussian is about half the lag of a Butterworth filters: Lag = N * P / (2 * ¶2), where,
N is the number of poles, and
P is the critical period
Special initialization of filter stages ensures proper working in scans with as few bars as possible.
From Ehlers Book: "The first objective of using smoothers is to eliminate or reduce the undesired high-frequency components in the eprice data. Therefore these smoothers are called low-pass filters, and they all work by some form of averaging. Butterworth low-pass filtters can do this job, but nothing comes for free. A higher degree of filtering is necessarily accompanied by a larger amount of lag. We have come to see that is a fact of life."
References John F. Ehlers: "Rocket Science For Traders, Digital Signal Processing Applications", Chapter 15: "Infinite Impulse Response Filters"
Included
Loxx's Expanded Source Types
Signals
Alerts
Bar coloring
Related indicators
STD-Filtered, Gaussian-Kernel-Weighted Moving Average
One-Sided Gaussian Filter w/ Channels
Fisher Transform w/ Dynamic Zones
R-sqrd Adapt. Fisher Transform w/ D. Zones & Divs.
Indicator
Indicator
Indicator
Indicator
Indicator
Indicator
Indicator
Nadaraya-Watson Smoothers [LuxAlgo]The following tool smoothes the price data using various methods derived from the Nadaraya-Watson estimator, a simple Kernel regression method. This method makes use of the Gaussian kernel as a weighting function.
Users have the option to use a non-repainting as well as a repainting method, see the USAGE section for more information.
🔶 USAGE
🔹 Non Repainting
When Repainting Smoothing is disabled the returned indicator acts similarly to a regular causal moving average. This result could be described as an "endpoint Nadaraya-Watson estimator".
Unlike a regular moving average whose degree of smoothness is commonly determined by the length of its calculation window, the degree of smoothness of the proposed indicator is determined by the bandwidth setting, with a higher value returning smoother results.
In the above chart, a bandwidth value of 50 is used. An increasing value of the smoother is indicative of an uptrend, while a decreasing value is indicative of a downtrend.
🔹 Repainting
Non-causal smoothing methods have found low support from technical analysts because they tend to repaint. Yet, they can provide powerful insights such as estimating underlying trends in the price as well as seeing how far prices deviate from them. They can also make drawing certain patterns easier and can help see underlying structures in the price more clearly.
Using higher bandwidth values allows for estimating longer-term trends in the price.
Triangular labels highlight points where the direction of the estimator change. This allows for the identification of tops and bottoms in the underlying trend which can be compared to the actual price tops and bottoms.
Note that multiple labels can appear in real time, highlighting real-time changes in the estimator's direction. The most recent label on a series of labels is the first to appear. This can eventually be useful for the real-time predictive application of the estimator. However, it is not a usage we particularly recommend.
🔶 DETAILS
The Nadaraya-Watson estimator can be described as a series of weighted averages using a specific normalized kernel as a weighting function. For each point of the estimator at time t , the peak of the kernel is located at time t , as such the highest weights are attributed to values neighboring the price located at time t .
A lower bandwidth value would contribute toward a more important weighting of the price at a precise point and would as such less smooth results. In the case where our bandwidth is so small that the resulting kernel is just an impulse, we would get the raw price back.
However, when the bandwidth is sufficiently large, prices would be weighted similarly, thus resulting in a result closer to the price mean.
It can be interesting to note that due to the nature of the estimator and its weighting procedure, real-time results would not deviate drastically for points in the estimator near the center of the calculation window.
🔶 SETTINGS
Bandwidth : controls the bandwidth of the Gaussian kernel, with higher values returning smoother results.
Src : Input source of the kernel regression.
Repainting Smoothing : Determine if the smoothing method should repaint or not. If disabled the "endpoint Nadaraya-Watson estimator" is returned. Indicator
Indicator
Resampling Filter Pack [DW]This is an experimental study that calculates filter values at user defined sample rates.
This study is aimed to provide users with alternative functions for filtering price at custom sample rates.
First, source data is resampled using the desired rate and cycle offset. The highest possible rate is 1 bar per sample (BPS).
There are three resampling methods to choose from:
-> BPS - Resamples based on the number of bars.
-> Interval - Resamples based on time in multiples of current charting timeframe.
-> PA - Resamples based on changes in price action by a specified size. The PA algorithm in this script is derived from my Range Filter algorithm.
The range for PA method can be sized in points, pips, ticks, % of price, ATR, average change, and absolute quantity.
Then, the data is passed through one of my custom built filter functions designed to calculate filter values upon trigger conditions rather than bars.
In this study, these functions are used to calculate resampled prices based on bar rates, but they can be used and modified for a number of purposes.
The available conditional sampling filters in this study are:
-> Simple Moving Average (SMA)
-> Exponential Moving Average (EMA)
-> Zero Lag Exponential Moving Average (ZLEMA)
-> Double Exponential Moving Average (DEMA)
-> Rolling Moving Average (RMA)
-> Weighted Moving Average (WMA)
-> Hull Moving Average (HMA)
-> Exponentially Weighted Hull Moving Average (EWHMA)
-> Two Pole Butterworth Low Pass Filter (BLP)
-> Two Pole Gaussian Low Pass Filter (GLP)
-> Super Smoother Filter (SSF)
Downsampling is a powerful filtering approach that can be applied in numerous ways. However, it does suffer from a trade off, like most studies do.
Reducing the sample rate will completely eliminate certain levels of noise, at the cost of some spectral distortion. The lower your sample rate is, the more distortion you'll see.
With that being said, for analyzing trends, downsampling may prove to be one of your best friends! Indicator
Right Sided Ricker Moving Average And The Gaussian DerivativesIn general gaussian related indicators are built by using the gaussian function in one way or another, for example a gaussian filter is built by using a truncated gaussian function as filter kernel (kernel refer to the set weights) and has many great properties, note that i say truncated because the gaussian function is not supposed to be finite. In general the gaussian function is represented by a symmetrical bell shaped curve, however the gaussian function is parametric, and the user might adjust the position of the peak as well as the width of the curve, an indicator using this parametric approach is the Arnaud Legoux moving average (ALMA) who posses a length parameter controlling the filter length, a peak parameter controlling the position of the peak of the gaussian function as well as a width parameter, those parameters can increase/decrease the lag and smoothness of the moving average output.
However what about the derivatives of the gaussian function ? We don't talk much about them and thats a pity because they are extremely interesting and have many great properties as well, therefore in this post i'll present a low lag moving average based on the modification of the 2nd order derivative of the gaussian function, i believe this post will be extremely informative and i hope you will enjoy reading it, if you are not a math person you can skip the introduction on gaussian derivatives and their properties used as filter kernel.
Gaussian Derivatives And The Ricker Wavelet
The notion of derivative is continuous, so we will stick with the term discrete derivative instead, which just refer to the rate of change in the function, we have a change function in pinescript, and we will be using it to show an approximation of the gaussian function derivatives.
Earlier i used the term 2nd order derivative, here the derivative order refer to the order of differentiation, that is the number of time we apply the change function. For example the 0 (zeroth) order derivative mean no differentiation, the 1st order derivative mean we use differentiation 1 time, that is change(f) , 2nd order mean we use differentiation 2 times, that is change(change(f)) , derivates based on multiple differentiation are called "higher derivative". It will be easier to show a graphic :
Here we can see a normal gaussian function in blue, its scaled 1st order derivative in orange, and its scaled 2nd derivative in green, note that i use scaled because i used multiplication in order for you to see each curve, else it would have been less easy to observe them. The number of time a gaussian function derivative cross 0 is based on the order of differentiation, that is 2nd order = the function crossing 0 two times.
Now we can explain what is the Ricker wavelet, the Ricker wavelet is just the normalized 2nd order derivative of a gaussian function with inverted sign, and unlike the gaussian function the only thing you can change is the width parameter. The formula of the Ricker wavelet is show'n here en.wikipedia.org , where sigma is the width parameter.
The Ricker wavelet has this look :
Because she is shaped like a sombrero the Ricker wavelet is also called "mexican hat wavelet", now what would happen if we used a Ricker wavelet as filter kernel ? The response is that we would end-up with a bandpass filter, in fact the derivatives of the gaussian function would all give the kernel of a bandpass filter, with higher order derivatives making the frequency response of the filter approximate a symmetrical gaussian function, if i recall a filter using the first order derivative of a gaussian function would give a frequency response that is left skewed, this skewness is removed when using higher order derivatives.
The Indicator
I didn't wanted to make a bandpass filter, as lately i'am more interested in low-lag filters, so how can we use the Ricker wavelet to make a low-lag low-pass filter ? The response is by taking the right side of the Ricker wavelet, and since values of the wavelets are negatives near the border we know that the filter passband is non-monotonic, that is we know that the filter will have low-lag as frequencies in the passband will be amplified.
So taking the right side of the Ricker wavelet only mean that t has to be greater than 0 and linearly increasing, thats easy, however the width parameter can be tricky to use, this was already the case with ALMA, so how can we work with it ? First it can be seen that values of width needs to be adjusted based on the filter length.
In red width = 14, in green width = 5. We can see that an higher values of width would give really low weights, when the number of negative weights is too important the filter can have a negative group delay thus becoming predictive, this simply mean that the overshoots/undershoots will be crazy wild and that a great fit will be impossible.
Here two moving averages using the previous described kernels, they don't fit the price well at all ! In order to fix this we can simply define width as a function of the filter length, therefore the parameter "Percentage Width" was introduced, and simply set the width of the Ricker wavelet as p percent of the filter length. Lower values of percent width reduce the lag of the moving average, but lets see precisely how this parameter influence the filter output :
Here the filter length is equal to 100, and the percent width is equal to 60, the fit is quite great, lower values of percent width will increase overshoots, in fact the filter become predictive once the percent width is equal or lower to 50.
Here the percent width is equal to 50. Higher values of percent width reduce the overshoots, and a value of 100 return a filter with no overshoots that is suited to act as a lagging moving average.
Above percent width is set to 100. In order to make use of the predictive side of the filter, it would be great to introduce a forecast option, however this require to find the best forecast horizon period based on length and width, this is no easy task.
Finally lets estimate a least squares moving average with the proposed moving average, you know me...a percent width set to 63 will return a relatively good estimate of the LSMA.
LSMA in green and the proposed moving in red with percent width = 63 and both length = 100.
Conclusion
A new low-lag moving average using a right sided Ricker wavelet as filter kernel has been introduced, we have also seen some properties of gaussian derivatives. You can see that lately i published more moving averages where the user can adjust certain properties of the filter kernel such as curve width for example, if you like those moving averages you can check the Parametric Corrective Linear Moving Averages indicator published last month :
I don't exclude working with pure forms of gaussian derivatives in the future, as i didn't published much oscillators lately.
Thx for reading !
Indicator
Roofing Filter [DW]This is an experimental study built on the concept of using roofing filters on price data proposed by John Ehlers.
Roofing filters are a type of bandpass filter conventionally used in HF radio receivers in the first IF stage to limit the frequency spectrum passed on to later stages in the receiver.
The goal in applying roofing filters to a price signal is to simultaneously attenuate high frequency noise and low frequency distortion to pass an oscillating signal with a nearly zero mean for analysis and/or further calculation.
In this study, there are three filter types to choose from:
-> Ehlers Roofing Filter, which passes data through a 2 pole high pass filter, then through a Super Smoother filter.
-> Gaussian Roofing Filter, which passes data through a 2 pole Gaussian high pass filter, then through a 2 pole Gaussian low pass filter.
-> Butterworth Roofing Filter, which passes data through a 2 pole Butterworth high pass filter, then through a 2 pole Butterworth low pass filter.
Each filter type has different amplitude and delay characteristics, so play around with each type and see which response suits your needs best.
There is an option to normalize the scale of the output as well. The normalization process in this script is computed by comparing positive and negative outputs to the filter's moving RMS value.
The resulting oscillator can be fed through numerous conventional indicators including Stochastic Oscillator, RSI, CCI, etc. to generate smoother, less distorted indicators for a clearer view of turning points.
Alternatively, it can also act as an indicator itself, as implied by the corresponding color scheme included in the script.
Although roofing filters are not conventionally used in the analysis of market data, applying such spectral analysis techniques may prove to be quite useful for the design of more efficient indicators and more reliable predictions. Indicator
Indicator
Strategy
Indicator
Indicator
Indicator
Indicator
Gaussian Channel [DW]This study is an experiment utilizing the Ehlers Gaussian Filter technique combined with lag reduction techniques and true range to analyze trend activity.
Gaussian filters, as Ehlers explains it, are simply exponential moving averages applied multiple times.
First, beta and alpha are calculated based on the sampling period and number of poles specified. The maximum number of poles available in this script is 9.
Next, the data being analyzed is given a truncation option for reduced lag, which can be enabled with "Reduced Lag Mode".
Then the alpha and source values are used to calculate the filter and filtered true range of the dataset.
Filtered true range with a specified multiplier is then added to and subtracted from the filter, generating a channel.
Lastly, a one pole filter with a N pole alpha is averaged with the filter to generate a faster filter, which can be enabled with "Fast Response Mode".
Custom bar colors are included.
Note: Both the sampling period and number of poles directly affect how much lag the indicator has, and how smooth the output is.
Larger inputs will result in smoother outputs with increased lag, and smaller inputs will have noisier outputs with reduced lag.
For the best results, I recommend not setting the sampling period any lower than the number of poles + 1. Going lower truncates the equation. Indicator
