optimal-execution

In this project, I started with a data extractor on binance. And I am happy to have contributors to implement a mock execution system. The objective is mainly educational so have fun!

Data Extraction

Once you've pulled the repo, create a folder data/json from the root of the project

mkdir -p data/json

To run the extraction try

go run cmd/dataextractor/dataextractor.go -symbol=BTCUSDT -json_path=data/json/ -throttle=5

If everything works, you should have prints on the screen like It provides an output like

symbol:  BTCUSDT json_path:  data/json/ throttle:  5
>>> Processed tick: 1
------------ Aggregated Data ------------
Last Timestamp: 1694867267234
Total Amount: 12831.961904
Total Number: 14
Total Volume: 0.483620
BuyPrice: 26533.150000
SellPrice: 26533.140000
Mid Price: 26533.145000
Spread: 0.007538
Order Size: 0.034544
Order Amount: 916.568707
Bid Total Amount: 263.208749
Bid Total Number: 1
Bid Total Volume: 0.009920
Bid Order Size: 0.009920
Bid Order Amount: 263.208749
Ask Total Amount: 12568.753155
Ask Total Number: 13
Ask Total Volume: 0.473700
Ask Order Size: 0.036438
Ask Order Amount: 966.827166
Bid Total Quantity: 371.889140
Ask Total Quantity: 254.197670
Bid Max Delta: 341.725000
Ask Max Delta: 316.845000
-----------------------------------------
------------ Order Book Data ------------
Last Update ID: 38853788103
Bids:
Price: 26533.14000000, Quantity: 2.17741000
Price: 26532.96000000, Quantity: 0.53350000
Price: 26532.95000000, Quantity: 0.04079000
Price: 26532.90000000, Quantity: 0.04085000
Price: 26532.85000000, Quantity: 0.03902000
Asks:
Price: 26533.15000000, Quantity: 5.53860000
Price: 26533.24000000, Quantity: 0.00200000
Price: 26533.63000000, Quantity: 0.10396000
Price: 26533.65000000, Quantity: 0.10396000
Price: 26533.67000000, Quantity: 0.10396000
-----------------------------------------
<<<
...

with prints on the screen every n=throttle seconds, as defined above. When you stop the extraction, a file is saved in data/json in the format capture_symbol=BTCUSDT_throttle=5_start=1694867262_end=1694867272.json where the timestamp will obviously be the ones of your start of the process and your end of it.

{
  "Spread": 0.007543766220855038,
  "OrderSize": 0.006236945827489481,
  "OrderAmount": 165.35367702397087,
  "BidOrderSize": 0.005939920430722318,
  "BidOrderAmount": 157.47893286249285,
  "BidTotalQty": 395.6825525000046,
  "BidMaxDelta": 327.0700000000006,
  "AskOrderSize": 0.0049890625,
  "AskOrderAmount": 132.269875328125,
  "AskTotalQty": 235.42801750000615,
  "AskMaxDelta": 338.02500000000146,
  "CumBids": [
    [
      0,
      0
    ],
    [
      0.0001,
      0.04905527602005688
    ],
    [
      0.0002,
      0.04934029523169036
    ],
    ...
  ],
  "CumAsks": [
    [
      0,
      0
    ],
    [
      0.0001,
      0.01546911719944717
    ],
    ...
  ]
}

The final output is what will provide you with the necessary data to compute the optimal execution.

If you wish to have a binary you can play with just do

go build -o bin/dataextractor cmd/dataextractor/dataextractor.go

if you do not have the bin folder yet a simple

mkdir bin

will solve that issue.

Pricer

It should compute for an Implementation Shortfall, the appropriate strategy in time and its value. As a reminder, the problem to solve is some minimisation of

$\left( q_0m_0 + \int_{0}^{q_0} F^{-1}\left(\chi\right)d\chi \right) - \dots - \left( q_{N - 1}m_{N - 1} + \int_{0}^{q_{N - 1}} F^{-1}\left(\chi\right)d\chi \right)$

with the constraint that $q_0 + \dots + q_{N - 1} = Q$.

Execution backtester

So here the task would be to try and get a feel for an actual task required as a quant trader: backtesting your strategy. I have done nothing yet, altough with the stream feed from binance, one could use that data to provide some estimation of the strategy performance (VWAP and TWAP as well if you're game!).

More on medium for instructions: link to medium