I've recently returned to using Jigsaw for my execution and immediately noticed that the depth of market on the Rithmic data feed is significantly lower (usually 30%) than on the Sierra Denali Feed. I was also able to compare Rithmic to CTS, and Rithmic again had a smaller depth of market pool.

I'm confident that the difference between the two will not have an impact on my executions, but I am intellectually curious as to why there is a such a discrepancy between the two.
Any ideas as to why the Rithmic depth of market is lower than Denali?

My best guess is that Rithmic must be using a pure MBO parser of MDP 3.0 so they are excluding implied depth from their book. This probably explains why it is strongly biased downwards from depths on the other feeds.

On the other hand Sierra and CTS probably using the (MBO +) MBP part of MDP 3.0 to aggregate the direct and implied depth.

The occasional times when Rithmic depth is higher are probably explained by packet loss (I recall at least one of these feeds uses UDP) or feed delay of either two feeds you're comparing between.

I usually use Rithmic's approach but there are situations where the other approach is useful

This is very strange, I did this comparison months ago and it was exactly the same.... are you sure you are comparing the same products? maybe you are just comparing micros vs minis, it looks very strange to me.

I did double check and both are on the ESH22 contract - I even double checked the Chart Settings in Sierra to confirm that "Trade and Current Quote Symbol" does not have the micro contract listed.

Do you use aggregated quotes in Rithmic?
You can check at the login screen.

So this isn't strictly to do with market depth, but I've discovered a major discrepancy between the two data feeds today which has me questioning which one is correct.

This is the CL open today 3rd February 2023 from approx. 8AM - 10AM central time.

Left is Motivewave running Rithmic non-aggregated data. Right is SC + Denali.
The bottom of both platforms is the cumulative delta. SC is using the cumulative delta - trades indicator.

As you can see, Motivewave/Rithmic is showing positive delta on the way up, SC/Denali is showing negative delta.
When I check the footprints on both on a random bar, the bids and asks and delta totals all add up on both.
When I compare the two feeds over the past few days, at a cursory glance, they're pretty much the same.

The fact that both are showing pretty much opposites of each other has me scratching my head.

Anyone else care to check and chime in?


SCR-20230204-1p7

artemiso 

My best guess is that Rithmic must be using a pure MBO parser of MDP 3.0 so they are excluding implied depth from their book. This probably explains why it is strongly biased downwards from depths on the other feeds.

I hate to question @artemiso but given that Rithmic seems to have 10 layers of market depth isn't it more likely to be MBP?

creamyyy 

So this isn't strictly to do with market depth, but I've discovered a major discrepancy between the two data feeds today which has me questioning which one is correct. This is the CL open today 3rd February 2023 from approx. 8AM - 10AM central time. Left is Motivewave running Rithmic non-aggregated data. Right is SC + Denali. The bottom of both platforms is the cumulative delta. SC is using the cumulative delta - trades indicator. As you can see, Motivewave/Rithmic is showing positive delta on the way up, SC/Denali is showing negative delta. When I check the footprints on both on a random bar, the bids and asks and delta totals all add up on both. When I compare the two feeds over the past few days, at a cursory glance, they're pretty much the same. The fact that both are showing pretty much opposites of each other has me scratching my head. Anyone else care to check and chime in? SCR-20230204-1p7

I'm not familiar with "cumulative delta" but gave it a shot.

It should be negative. Values below differ because your start time and mine were different (I used exactly 08:00 CT, I think your cumulative sum started around 09:00+). I tried 00:00 UTC and got a similar negative curve in any case.



Python script below in case you want to try.

 

Code

import databento as db import pandas as pd import matplotlib.pyplot as plt plt.style.use('ggplot') client = db.Historical(key='YOUR_API_KEY') data = client.timeseries.stream(dataset='GLBX.MDP3', schema='trades', stype_in='smart', # use smart lead month symbol symbols=['CL.v.0'], # lead month CL by volume start='2023-02-03', end='2023-02-04') # Convert to dataframe with UNIX timestamps and display prices df = data.to_df(pretty_ts=False, pretty_px=True) # Convert UNIX timestamps to US Central Time df.index = pd.to_datetime(df.index).tz_localize('UTC').tz_convert('America/Chicago') df['signed_vol'] = df['side'].apply(lambda s: 1 if s == 'B' else -1) * df['size'] df = df.between_time('08:00', '10:00') df['cumulative_delta'] = df['signed_vol'].cumsum() plt.plot(df['cumulative_delta']) plt.xlabel('Time (Chicago)') plt.ylabel('cumusum(signed_vol)') plt.show()

SMCJB 

I hate to question @artemiso but given that Rithmic seems to have 10 layers of market depth isn't it more likely to be MBP?

Your hypothesis is reasonable, but keep in mind it's possible to:

• Build the book on MBO and then disseminate MBP. ("Correct" way to do it, IMO.)
• Pass on MBP directly.

I don't have Rithmic. The easiest to tell if you're right is to subscribe to a 3 or 5 deep book instrument, say GE.

Reason I suspected the implied book is because it's quite natural to ignore implied depth when processing MBO since the implied book is disseminated via MBP. On revisiting this issue, I realized the OP is seeing this discrepancy on every level though, which rules out this being purely an implied book issue. The Denali feed actually looks suspicious too here because the book depth looks too uniform across every level. So two other possibilities that might even be happening concurrently:

• Maybe the Denali side is showing the 'last seen implied depth' at the level without deleting it after it goes out of scope?
• Maybe Rithmic is parsing MBO and because their infrastructure is lossy, they're missing a lot of messages and hence missing many adds, and they just send whatever they've seen.

The fastest way to tell is if OP can share the time of this observation.

Kind of scary either way that two DOM's could be so different!

I've never done it before but I have to data providers so may put them up next to each other to compare when I'm back at home.