Working out Value Area (a.k.a Market Profile/Volume Profile from Auction Market Theory)

Still working on this but I think I know what the issue is...the 'comparing volume of price intervals' starts from the highest volume interval and works its way out, but isn't robust to local minimas, so it'll gets stuck in an infinite loop if the volume profile doesn't follow a nice bell shape e.g on the example chart here https://tlc.thinkorswim.com/center/reference/Tech-Indicators/studies-library/V-Z/VolumeProfile

Will update this as I try other methods...

I am very interested in this; if you are a 'pro' on tradingview.com you also get access to the 'value area' metrics which are very useful! 

I've made a working script in Python that calculates volume-based POC as well as time-based POC. However it runs quite slowly as is apparent when calculating over large datasets (1H history over 2 years of data takes ~8 minutes on a single symbol!). I'd be interested in hearing of an implementation that somehow avoids "binning" volume/time @ price but instead relies only on a novel idea for accumulating/decumulating volume/time @ price that saves on memory/iterations.

I was mulling over a notion of creating a "vector path histogram", where the beginning is the lowest price encountered (or just simply $0), and the end point is the highest value encountered. The step-wise curve itself is defined by the accumulation of bar data as it's encountered and decumulation as it passes beyond the lookback period.

So when you encounter new bar data, you create two 90 degree pivot anchors for each high/low when/where required which rises/fells the histogram line by the quantity you specify (volume / time). So the only object retained in memory is a vector path with a lower limit of 0 and an unbound upper limit. However I don't know how to program this or what the optimal mathematics is.

Jake Mitchell , happy to take a look at optimizing your data vector. It sounds like a great idea.

Hi everyone,

I've attached a Python backtest which implements a technique for calculating the Point of Control for a given day. We can accomplish this by setting our universe resolution to minute/second and determining the volume traded at each level over the intraday candles in OnData. When the day is over, we can build a window which captures the prices levels and the associated volumes traded. We start this window at the price level which had the most volume throughout the day and grow it incrementally in the direction of the price level that had the most volume (up or down) until we have 70% of the day's volume captured.

See the attached backtest and code for a full working example.

Best,
Derek

Just wanted to comment here on the performance issue. I've found that migrating some of our algos to C# improved speeds by nearly 50X! We had backtests in Python that took ~6min per symbol to process, but that completed in just 8 seconds in C#. We're now writing all our algorithms in C#, and only using Python during research within the Python notebook Research environment. Hope this helps!

Hi Kbsaravana,

This related thread extends the use of this indicator.

Best,
Derek Melchin

Derek Melchin : interested in this. any chance you have an updated link? The one you shared leads to a 404. 

Fixed ekz

Hi Derek,

Thank you for sharing your code. I was trying to create a program that would take the symbol and date as input and spit out the POC for the day. I was trying to use your solution as the base and ran into couple of issues:

a). I tried running it for 11th and 12th of Nov and it gives the POC as 462.82. On 11th at 12th the price does not touch 462.

b). Is it possible to simply take a single date as start date and just aggregate volume for that date?

c).Does QC allow file IO? I want to write the output to a file. I searched the documentation and ran into custom reader but nothing on writing output.

 Any help would be greatly appreciated.

Thanks,

Prakash

Hi Prakash Murali,
Would you mind sharing the backtest you are working with for a precise diagnosis?
No, we cannot do file IO it's a requirement of our data vendors to avoid data theft.
We recommend using the ObjectStore for storing or retrieving data during backtesting or the research environment. Refer to the following thread for more information.
Best,
Varad Kabade