One of the most sought after features in the NeuroShell Power User versions is the ability to mix timeframes on a single chart. For example, you can have an indicator based on hourly data as part of trading rules or predictions in a 15 minute chart and possibly prevent unnecessary trades when the data is noisy.
The chart below shows two different methods of using hourly data to calculate the RSI indicator on a 15 minute chart. The subgraph that contains blue and yellow dots displays the difference. We’ll discuss how you can set up the indicator so it calculates correctly for your purposes.
The Setting at the Beginning of the Wizard
Blue and yellow dots display different versions of RSI created with hourly bars.
In the 15 minute chart for SPY, we have added two RSI indicators that use hourly bars in the calculations. The RSI blue dot indicator is calculated once an hour. The RSI yellow dot indicator is calculated every 15 minutes and repeats the same value for 4 bars. The difference is the setting that appears at the beginning of the indicator wizard.
Let’s take a closer look.When you start the indicator wizard, you are asked to “Select the indicator frequency”. This setting controls how often the indicator is calculated. If you select Option A below, “Same frequency as chart”, on a 15 minute chart the indicator will be calculated every 15 minutes. In our SPY chart, this is the setting we used for the yellow dot version of the RSI indicator. You’ll see four yellow dots that display the same value, which is correct because the indicator is using 1 hour bars and the RSI value remains constant for the entire hour.
Option A: Indicator frequency same as chart created the yellow dots for the RSI indicator
Option B: Indicator frequency independent of chart produced the blue dots for RSI
If you choose Option B for Chart Independent Frequency for 1 hour bars, the indicator will only be calculated once an hour and will be displayed like the blue dots on our Spy chart.
Step 2: The Data Stream In Step 2 of the Indicator Wizard you choose the data stream for the closing price used to calculate RSI. In both options the Closing Price was set to the 1 Hour Close. Why do the two indicators appear at different levels on the same bar?
The RSI indicator calculations include a moving average. In the yellow dot version that is calculated every 15 minutes there are four duplicate values for each hour of training data included in the average. In the blue dot version, which is only calculated once an hour, there are many different values included in the average. The result can be different RSI values for the same bar.
Which one is correct? If you’re are trying to include a longer term view of the market in your trading decisions, you will probably obtain more reliable information from the RSI indicator that is calculated once an hour (the blue dots). The second method of changing the frequency of the input may not make as much sense when using the RSI, however it does open a world of possibilities when using multiple input indicators like arithmetic and rule indicators where you can compute, compare and contrast values from different time frames. In the end, the decision is up to you.
The Prediction and Trading Strategy Wizards include the same chart independent frequency options.
BUILDING TRADING MODELS THAT WORK IN THE FUTURE IN CHAOSHUNTER AND NEUROSHELL TRADER
We hear from customers who express their concern that the profitable models they build don’t hold up in live trading. Here are several techniques you can try that will enhance the likelihood of success in the future.
If you are using an optimization goal function in the trading strategy section of ChaosHunter or any Trading Strategy or Prediction in Trader, we suggest that you increase commissions so the model will only look for big moves. Larger commissions will make the model view trades with small profits as losers and look for trades that generate more profit.
Another technique is to split your data into a training and out-of-sample set on the data grid using the Select Ranges button in ChaosHunter and the Dates tab in NeuroShell Trader. That way you can examine the results for both the training and out-of-sample data. Save the model that is generating trading signals at the peaks and valleys of both data sets.
One final thought is to build an ensemble of three models and buy when two out of the three are producing a signal. That will increase the probability that you are getting a correct signal.
If you want more information on how much data to use for training a model and how much out-of-sample data to use, click here to read “Selecting in-sample, paper, and out-of-sample periods” commentary by Steve Ward.
Remember, no model will last forever and you have to be vigilant about retraining or changing models as market conditions evolve.
