Forecast OscillatorThe Forecast Oscillator is a technical indicator that compares a security close price to its time series forecast. The time series forecast function name is "tsf" and it calculates the projection of the price trend for the next bar.
The Forecast Oscillator and therefore the time series forecast are based on linear regression. The time series forecast indicator is equal to the sum of two other indicators: the linear regression (LinearReg) and the linear regression slope (LinearReg_Slope).
If the Forecast Oscillator stays above the zero line for an extended period, then it signals that the price may rise in the future and if it stays below the zero line for an extended period, then it signals a coming fall in the security price.
The indicator name is "Forecast Oscillator" and it accepts two arguments. The first argument is the time series that is used in the next bar forecast (It is usually the close price) and the second one is the period that will be passed to the time series forecast function during calculation . The technical analysis indicator returns a value in percentage that corresponds to the close price minus the previous value of the time series forecast, multiplied by 100 and divided by the close price.
Author's Note:
Just look at the exaggerated movements of the oscillator especially in trend changes . Some examples can be experienced on the chart in rectangles.
Kıvanç Özbilgiç
Forecasting
Garch (1,1) ModelThe Garch (General Autoregressive Conditional Heteroskedasticity) model is a non-linear time series model that uses past data to forecast future variance.
The Garch (1,1) formula is:
Garch = (gamma * Long Run Variance) + (alpha * Squared Lagged Returns) + (beta * Lagged Variance)
The gamma, alpha, and beta values are all weights used in the Garch calculations. According to RiskMetrics by JP Morgan, the optimal beta weight is 0.94, but this figure is highly disputed in the academic realm. The biggest problem academics and economists have with the 0.94 figure is that JP Morgan used monthly data to come to this number, meaning it does not take other time frames into account. Because of the disputed nature of what beta should be, this script will automatically calculate the beta weight for you in real time, taking into account the time frame you're using and realized variance, by using the Minimum Sum of Squared Errors Method.
The gamma and alpha weights are also calculated for you.
Even though the Garch formula provides today's projected variance, today's projected deviation is also calculated. This is done by taking the square root of Garch.
Additionally, if you want to project the variance or deviation for as many days forward as you want, you can.
In order to project the variance and deviation beyond just today, these equations are used:
Projected Variance = Long Run Variance + (alpha + beta)^Days Forward * (Garch - Long Run Variance)
Projected Deviation = sqrt(Projected Variance)
How to use this model:
1st. Decide the type of data you want: Projected Variance in % or Projected Deviation in %.
2nd. Decide how many days you want projected forward. If you input 0, you will get projections for today. If you input 1, you will get projections for tomorrow, and etc.
That's it. If you have any further questions, I left detailed comments in the code explaining each step, as best as I could.
MultiTimeFrame Shifting Predictive Bollinger BandsThis is the optimized version of my MTFSBB indicator with capability of possible bands prediction in case of negative shifting (to the left).
Make me happy by using it and sending me your ideas about the prediction.
Grand Trend Forecasting - A Simple And Original Approach Today we'll link time series forecasting with signal processing in order to provide an original and funny trend forecasting method, the post share lot of information, if you just want to see how to use the indicator then go to the section "Using The Indicator".
Time series forecasting is an area dealing with the prediction of future values of a series by using a specific model, the model is the main tool that is used for forecasting, and is often an expression based on a set of predictor terms and parameters, for example the linear regression (model) is a 1st order polynomial (expression) using 2 parameters and a predictor variable ax + b . Today we won't be using the linear regression nor the LSMA.
In time series analysis we can describe the time series with a model, in the case of the closing price a simple model could be as follows :
Price = Trend + Cycles + Noise
The variables of the model are the components, such model is additive since we add the component with each others, we should be familiar with each components of the model, the trend represent a simple long term variation of high amplitude, the cycles are periodic fluctuations centered around 0 of varying period and amplitude, the noise component represent shorter term irregular variations with mean 0.
As a trader we are mostly interested by the cycles and the trend, altho the cycles are relatively more technical to trade and can constitute parasitic fluctuations (think about retracements in a trend affecting your trend indicator, causing potential false signals).
If you are curious, in signal processing combining components has a specific name, "synthesis" , here we are dealing with additive synthesis, other type of synthesis are more specific to audio processing and are relatively more complex, but could be used in technical analysis.
So what to do with our components ? If we want to trade the trend, we should estimate right ? Estimating the trend component involve removing the cycle and noise component from the price, if you have read stuff about filters you should know where i'am going, yep, we should use filters, in the case of keeping the trend we can use a simple moving average of relatively high period, and here we go.
However the lag problem, which is recurrent, come back again, we end up with information easier to interpret (here the trend, which is a simple fluctuation such as a line or other smooth curve) at the cost of decision timing, that is unfortunate but as i said the information, here the moving average output, is relatively simple, and could be easily forecasted right ? If you plot a moving average of high period it would be easier for you to forecast its future values. And thats what we aim to do today, provide an estimate of the trend that should be easy to forecast, and should fit to the price relatively well in order to produce forecast that could determine the position of future closing prices observations.
Estimating And Forecasting The Trend
The parameter of the indicator dealing with the estimation of the trend is length , with higher values of length attenuating the cycle and noise component in the price, note however that high values of length can return a really long term trend unlike a simple moving average, so a small value of length, 14 for example can still produce relatively correct estimate of trend.
here length = 14.
The rough estimate of the trend is t in the code, and is an IIR filter, that is, it is based on recursion. Now i'll pass on the filter design explanation but in short, weights are constants, with higher weights allocated to the previous length values of the filter, you can see on the code that the first part of t is similar to an exponential moving average with :
t(n) = 0.9t(n-length) + 0.1*Price
However while the EMA only use the precedent value for the recursion, here we use the precedent length value, this would just output a noisy and really slow output, therefore in order to create a better fit we add : 0.9*(t(n-length) - t(n-2length)) , and this create the rough trend estimate that you can see in blue. On the parameters, 0.9 is used since it gives the best estimate in my opinion, higher values would create more periodic output and lower values would just create a rougher output.
The blue line still contain a residual of the cycle/noise component, this is why it is smoothed with a simple moving average of period length. If you are curious, a filter estimating the trend but still containing noisy fluctuations is called "Notch" filter, such filter would depending on the cutoff remove/attenuate mid term cyclic fluctuations while preserving the trend and the noise, its the opposite of a bandpass filter.
In order to forecast values, we simply sum our trend estimate with the trend estimate change with period equal to the forecasting horizon period, this is a really really simple forecasting method, but it can produce decent results, it can also allows the forecast to start from the last point of the trend estimate.
Using The Indicator
We explained the length parameter in the precedent section, src is the input series which the trend is estimated, forecast determine the forecasting horizon, recommend values for forecast should be equal to length, length/2 or length*2, altho i strongly recommend length.
here length and forecast are both equal to 14 .
The corrective parameter affect the trend estimate, it reduce the overshoot and can led to a curve that might fit better to the price.
The indicator with the non corrective version above, and the corrective one below.
The source parameter determine the source of the forecast, when "Noisy" is selected the source is the blue line, and produce a noisy forecast, when "Smooth" is selected the source is the moving average of t , this create a smoother forecast.
The width interval control...the width of the intervals, they can be seen above and under the forecast plot, they are constructed by adding/subtracting the forecast with the forecast moving average absolute error with respect to the price. Prediction intervals are often associated with a probability (determining the probability of future values being between the interval) here we can't determine such probability with accuracy, this require (i think) an analysis of the forecasting distribution as well as assumptions on the distribution of the forecasting error.
Finally it is possible to see historical forecasts, that is, forecasts previously generated by checking the "Show Historical Forecasts" option.
Examples
Good forecasts mostly occur when the price is close to the trend estimate, this include the following highlighted periods on AMD 15TF with default settings :
We can see the same thing at the end of EURUSD :
However we can't always obtain suitable fits, here it is isn't sufficient on BTCUSD :
We can see wide intervals, we could change length or use the corrective option to get better results, another option is to use a log scale.
We will end the examples with the log SPX, who posses a linear trend, so for example a linear model such as a linear regression would be really adapted, lets see how the indicator perform :
Not a great fit, we could try to use an higher length value and use "Smooth" :
Most recent fits are quite decent.
Conclusions
A forecasting indicator has been presented in this post. The indicator use an original approach toward estimating the trend component in the closing price. Of course i should have given statistics related to the forecasting error, however such analysis is worth doing with better methods and in more advanced environment allowing for optimization.
But we have learned some stuff related to signal processing as well as time series analysis, seeing a time series as the sum of various components is really helpful when it comes to make sense of chaotic and noisy series and is a basic topic in time series analysis.
You can see that in this new year i work harder on the visual of my indicators without trying to fall in the label addict trap, something that i wasn't really doing before, let me know what do you think of it.
Thanks for reading !
Scripting Tutorial B - TManyMA - Commission/FeesThis script is for a triple moving average strategy where the user can select from different types of moving averages, price sources, lookback periods and resolutions.
Features:
- 3 Moving Averages with variable MA types, periods, price sources, resolutions and the ability to disable each individually.
- Crossovers are plotted on the chart with detailed information regarding the crossover (Ex: 50 SMA crossed over 200 SMA )
- Forecasting available for all three MAs. MA values are forecasted 5 values out and plotted as if a continuation to the MA.
- Forecast bias also applies to all forecasting. Bias means we can forecast based on an anticipated bullish , bearish or neutral direction in the market.
- To understand bias, please read the source code, or if you can't read the code just send me a message on here or Twitter . Twitter should be linked to my profile.
- Ribbons added and on by default. Optional setting to disable the ribbons. 5 ribbons between MA1 and MA2 and another 5 between MA2 and MA3.
- Ribbons are alpha-color coded based on their relation to their default MAs.
- Ribbons are only visible between MAs if the MAs being compared share the same Type, Resolution, and Source because there is no way to consolidate those three in a simple manner.
- Ribbon values are calculated based on calculated MA Periods between the MAs.
- Converted the existing study into a strategy.
- Strategy only enters long positions with a market order when MA crossovers occur.
- Strategy exits positions when crossunders occur.
- Trades 100% of the equity with one order/position by default.
- Ability to disable trading certain crosses with input checks.
- Ability to exit trades with a take profit or stop loss.
- User input to allow quick changes to the take profit or stop loss percentages.
- Strategy now calculates on every tick
- Strategy also includes fixed commission values based on Coinbase standard order fees
This script is meant as an educational script with well-formatted styling, and references for specific functions.
*** PLEASE NOTE - THIS STRATEGY IS MEANT FOR LEARNING PURPOSES. DEPENDING ON IT'S CONFIGURATION IT MAY OR MAY NOT BE USEFUL FOR ACTUAL TRADING. THE STRATEGY IS NOT FINANCIAL ADVICE ***
Scripting Tutorial 9 - TManyMA Strategy - Long Market Order OnlyThis script is for a triple moving average strategy where the user can select from different types of moving averages, price sources, lookback periods and resolutions.
Features:
- 3 Moving Averages with variable MA types, periods, price sources, resolutions and the ability to disable each individually
- Crossovers are plotted on the chart with detailed information regarding the crossover (Ex: 50 SMA crossed over 200 SMA )
- Forecasting available for all three MAs. MA values are forecasted 5 values out and plotted as if a continuation to the MA.
- Forecast bias also applies to all forecasting. Bias means we can forecast based on an anticipated bullish, bearish or neutral direction in the market.
- To understand bias, please read the source code, or if you can't read the code just send me a message on here or Twitter. Twitter should be linked to my profile.
- Ribbons added and on by default. Optional setting to disable the ribbons. 5 ribbons between MA1 and MA2 and another 5 between MA2 and MA3.
- Ribbons are alpha-color coded based on their relation to their default MAs.
- Ribbons are only visible between MAs if the MAs being compared share the same Type, Resolution, and Source because there is no way to consolidate those three in a simple manner.
- Ribbon values are calculated based on calculated MA Periods between the MAs.
- Converted the existing study into a strategy
- Strategy only enters long positions with a market order when MA crossovers occur
- Strategy exits positions when crossunders occur
- Trades 100% of the equity with one order/position by default
- Ability to disable trading certain crosses with input checks
This script is meant as an educational script with well-formatted styling, and references for specific functions.
*** PLEASE NOTE - THIS STRATEGY IS MEANT FOR LEARNING PURPOSES. DEPENDING ON IT'S CONFIGURATION IT MAY OR MAY NOT BE USEFUL FOR ACTUAL TRADING. THE STRATEGY IS NOT FINANCIAL ADVICE ***
Scripting Tutorial 7 - Triple Many Moving Averages ResolutionsThis script is for a triple moving average indicator where the user can select from different types of moving averages, price sources, lookback periods and resolutions.
Features:
- 3 Moving Averages with variable MA types, periods, price sources, resolutions and the ability to disable each individually
- Crossovers are plotted on the chart with detailed information regarding the crossover (Ex: 50 SMA crossed over 200 SMA )
- Forecasting available for all three MAs. MA values are forecasted 5 values out and plotted as if a continuation to the MA.
- Forecast bias also applies to all forecasting. Bias means we can forecast based on an anticipated bullish, bearish or neutral direction in the market.
- To understand bias, please read the source code, or if you can't read the code just send me a message on here or Twitter. Twitter should be linked to my profile.
This script is meant as an educational script with well-formatted styling, and references for specific functions.
Scripting Tutorial 6 - Triple Many Moving Averages ForecastingThis script is for a triple moving average indicator where the user can select from different types of moving averages, price sources and lookback periods.
Features:
- 3 Moving Averages with variable MA types, periods, price sources and ability to disable each individually
- Crossovers are plotted on the chart with detailed information regarding the crossover (Ex: 50 SMA crossed over 200 SMA )
- Forecasting available for all three MAs. MA values are forecasted 5 values out and plotted as if a continuation to the MA.
- Forecast bias also applies to all forecasting. Bias means we can forecast based on an anticipated bullish, bearish or neutral direction in the market.
- To understand bias, please read the source code, or if you can't read the code just send me a message on here or Twitter. Twitter should be linked on my profile.
This script is meant as an educational script with well-formatted styling, and references for specific functions.
Forecasting - Quadratic RegressionThis script is written totally thanks to Alex Grover (). Here it is implemented in conjunction with the seasonal forecast I showed in one of my previous posts. It takes the calculated QReg curve and extends its last section (Season) into the future (Forecasted periods).
Forecasting - Seasonal Naive MethodFor completeness here is a naive method with seasonality. The idea behind naive method with seasonality is to take last value from same season and treat it as a forecast. Its counterpart, naive method without seasonality, involves taking last mean value, i.e forecast = sma(x, p) .
Forecasting - Simple Mean MethodThis is a continuation of my series on forecasting techniques. The idea behind the Simple Mean method is to somehow extend historical mean to the future. In this case a forecast equals to last value plus average change.
Forecasting - Locally Weighted Regression (rescaled)UPDATE: the original version works only with BTC. Here's a general version with rescaling.
Forecasting - Vanilla Locally Weighted RegressionThere is not much to say - just vanilla locally weighted regression in PineScript 4.
see: medium.com
also: cs229.stanford.edu
Forecasting - Least Squares RegressionTested on 5m TF with EURUSD. Settings should be modified appropriately for other TFs, lookbacks and securities. This indicator does not repaint.
Forecasting - Holt’s Linear Trend ForecastingHolt's Forecasting method
Holt (1957) extended simple exponential smoothing to allow the forecasting of data with a trend. This method involves a forecast equation and two smoothing equations (one for the level and one for the trend):
Forecast equation: ŷ = l + h * b
Level equation: l = alpha * y + (1 - alpha) * (l + b)
Trend equation: b = beta * (l - l) + (1 - beta) * b
where:
l (or l) denotes an estimate of the level of the series at time t,
b (or b) denotes an estimate of the trend (slope) of the series at time t,
alpha is the smoothing parameter for the level, 0 ≤ alpha ≤ 1, and
beta is the smoothing parameter for the trend, 0 ≤ beta ≤ 1.
As with simple exponential smoothing, the level equation here shows that l is a weighted average of observation y and the one-step-ahead training forecast for time t, here given by l+b. The trend equation shows that b is a weighted average of the estimated trend at time t based on l-l and b, the previous estimate of the trend.
The forecast function is not flat but trending. The h-step-ahead forecast is equal to the last estimated level plus h times the last estimated trend value. Hence the forecasts are a linear function of h.
Stationary Extrapolated Levels OscillatorIntroduction
The oscillator version of the stationary extrapolated levels indicator. The methodology behind the extrapolated levels where to minimize the risk of making a decision based only on a forecast, therefore the indicator plotted levels in order to determine possible reversal points, signals where generated when the detrended series crossed over/under those levels.
The Indicator
First we detrend the price, this is because forecasting the trend is often harder than a series without trend (stationarity > non-stationarity) , then we forecast the detrended price with a linear extrapolation over a period of length and apply a max/min filter twice to the forecast, the max/min filters are just the highest and lowest function in pine. So the max/min filter have lag length/2 , by applying it two times we have a lag of length which is the period of the forecast. Because we use highest and lowest we can apply min-max normalization in the form of :
x' = (x - min(x, min'))/(max(x,max') - min(x, min'))
where x is the detrended price, max' the highest of the forecast of x and min' the lowest of the forecast of x . This result in a scaled oscillator in a range of (1,0),
When the indicator is equal to 1 or 0 there are high chances of reversals, more in depth this mean that the detrended price have crossed the highest/lowest of the forecast, when the indicator is equal to 0 or 1 for a long time this mean that the forecast was quite inaccurate, you can minimize risk by focusing on the cross between the detrended price and the 0.8/0.2 levels.
Conclusion
I've shown an oscillator version of my previous "Stationary extrapolated levels" indicator, the method involving taking the highest and lowest of the forecast is a great way to minimize the risk involved by time-series forecasting driven decisions. So i hope you find an use to it.
Thanks for reading !