Skip to content
/ Time-priority-and-risk-sharing Public

Code and data example for Garriott, van Kervel, and Zoican (2022). The data is randomly generated, while preserving the structure of the proprietary Montreal Exchange data.

0 stars 0 forks Branches Tags Activity
Star
Notifications

mariuszoican/Time-priority-and-risk-sharing

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits

.idea

.idea

 
 

Code

Code

 
 

Output

Output

 
 

ProcessedData

ProcessedData

 
 

RegressionCode

RegressionCode

 
 

TMXData

TMXData

 
 

.gitattributes

.gitattributes

 
 

readme.md

readme.md

 
 

readme_data.md

readme_data.md

 
 

Repository files navigation

Workflow for TMX project

1. Market-maker labels and risk-sharing results.

  • All the Python codes below can be run (in order) from batch file Code\0_batch_inventory.py

1.1 Generate market-maker labels

  1. Run generate_list_traders.py to get a list of trading accounts. Saves resulting panel at ProcessedData\account.csv.
  2. Run account_dailysummary.py to obtain a trader-stock-day panel with Kirilenko, Kyle, Samadi and Tuzun (2017, JF) variables -- net position, number of trades/volume traded, standard deviation of inventory. Saves a panel at ProcessedData\account_EODdata.csv.
  3. Run BBOpresence.py to obtain a panel with each traders' time share at best bid and ask (or both ask and bid). Saves the panel at ProcessedData\BBO_presence_MM.csv.
  4. Run mm_hft_label.py to assign labels for market makers (using KKST-based methodology and presence at the BBO) as well as HFT labels based on TMX information. Saves file with labels at ProcessedData\mm_hft_labels.csv.

1.2 Study market-maker inventories

  1. Run inventory_panel.py to get inventory panels for each market maker at various frequencies (e.g., 30 second, 1 minute, 30 minutes, 1 hour). Panels are saved in ProcessedData/Inventories/inventory_panel_FREQUENCY.csv.
  2. Run mean_reversion_inventory.py to estimate inventory half-life (in minutes) for each market-maker using an AR(1) process. Half-life panel saved as ProcessedData/inventory_halflife.csv.
  3. Run mean_reversion_graph.py to generate a two-panel figure of BBO presence and inventory half-lives across market makers. Figure is saved as Output/BBOshares_Halflife_MM.png.

1.3 Study risk-sharing by market-makers

  1. Run risksharing_correlations.py to obtain the average pairwise correlation between MM inventory based on 30-second inventory panel. Output saved as ProcessedData/Inventories/InvCorrelations.csv.
  2. Run risksharing_inefficiency.py to obtain estimates of MM risk-sharing inefficiency (based on a quadratic inventory model). Inefficiency panel saved as ProcessedData/Inventories/InvInefficiency.csv.
  3. Run risksharing_graphs.py to plot the average pairwise correlation and risk-sharing inefficiency. Output saved as Output/mm_risksharing.png

2. Generate the snapshot-trader marginal quote panel and snapshot depth panel.

  • All the codes below can be run (in order) from batch file Code\1_batch_getmainpanel.py.
  1. Run build_panel.py on the raw data to generate a folder with marginal quote panels for all traders (ProcessedData\MarginalQuotePanels\). The code uses auxiliary functions from functions_tmxdata.py.
  2. Run fill_zero_quant.py to take each marginal quote panel and:
    1. Assign quantities of zero on the no-quote side for traders active on one side of the book. Such zero-quantities enter the book "as if" they are at the lowest priority.
    2. Merge the trader IDs with the market-maker or HFT label.
    3. Resample the panels every 30s (to avoid double-counting quotes)
    4. Save the resulting panels in ProcessedData\MarginalQuotePanels_ZF\
  • Running fill_zero_quant.py takes a lot of time. The folder Code\ForSlurm\ includes a slightly modified version and a SLURM script to run the code in parallel on the Rotman Research Node. Panels need to be manually transferred to the repo, using e.g., WinSCP.
  1. Run select_mm_data.py to only select the market-maker quotes and save to file ProcessedData\mquotes_ZF_mm.csv. This is the main panel for regressions.
  2. Run depth_processing.py to build a panel with 30-second order book snapshots and correlation coefficients between inventory and queue position. Save file as ProcessedData\depth_snapshots.csv. The file also generates two representative figures: Output/depth_correlation.png (impact of inventory-queue correlation on depth) and Output/rho_distribution.png (distribution of inventory-queue correlations).
  3. Use sumstats.py to generate the figures for adverse selection (Output/queue_quantities.png) and inventory (Output/inventory_concerns.png).
  4. Run mm_quotes_preliminaries.py to generate mquotes_mm.csv which is the same as mquotes_ZF_mm.py but without zero quantities filled.
  5. Run pivot_snapshots.py to obtain a panel with marginal inventories for MMs as a function of their relative priority level in the book. Panel is saved at ProcessedData/pivot_quotes_inventories.csv.

3. Econometric analysis

  1. Use RegressionCode/summary_stats.R to produce summary stats tables for market-makers (Output/Tables/mmstats.tex), non-market makers (Output/Tables/nonmmstats.tex) and for all quote snapshots (Output/Tables/sumstats.tex)
  2. Use RegressionCode/regression_tests.R for econometric analysis on mquotes_ZF_mm.csv. Output is a table with structural regression estimates: Output/Tables/ASICtable_main.tex
  3. Use RegressionCode/depth_tests.R for econometric analysis on depth_snapshots.csv. Output is a table with estimates on depth effects: Output/Tables/Depth_main.tex.
  4. Use RegressionCode/inventory_order_depth.R for econometric analysis on pivot_quotes_inventories.csv. Output is a table with the marginal impact of inventory as a function of queue position: Output/Tables/Depth_Ordering.tex.

4. Miscellenea

  1. Use theory_figure.py to generate theory-supporting figures Output/theory_example_params.png and Output/marginal_impact_queue.png.
  2. Use get_average_trade.py to estimate the average trade size in the data.

About

Code and data example for Garriott, van Kervel, and Zoican (2022). The data is randomly generated, while preserving the structure of the proprietary Montreal Exchange data.

Topics

research-paper

Resources

Readme
Activity

Stars

0 stars

Watchers

2 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

No packages published

Languages