[Admin] Meeting on 14.03.18

[mjuchli]

Agenda

1) Research questions (material A, B, C):

• rule out possible pitfalls and/or ambiguities
• make sure we rise important/valid questions
• being able to (re)define the questions precisely

2) Process/Pipeline (material E)

• discuss the process such that I can work future months really targeted
• and make sure it will cover the defined questions

3) Patterns in orders (material D)

• reason about an attempt to classify the impact on order execution based on past order/trade behaviour

4) Schedule talk

Material

A) Understand the basic statistical definitions of the order execution problem: http://discovery.ucl.ac.uk/1359852/1/Chaiyakorn%20Yingsaeree%20-%20Thesis.pdf

• Page 179-180: 7.3 Framework for an order placement strategy
• Page 185-187: 7.4.3 Empirical unconditional model using density estimation (I think those 4 pages define the difficulties of order execution clearly and is therefore a good starting point to remind ourselves what the context really is about).

B) Roughly be able to follow the demonstrated order executions on the artificial order books: Section "Expected Execution on artificial prices" in https://github.com/backender/ctc-executioner/blob/master/notebooks/order_execution_behaviour.ipynb e.g. understand that

• BUY orders can be placed lower when the price falls
• BUY orders must be placed higher when the price rises
• SELL orders can be placed higher when the price rises
• SELL orders must be placed lower when the price falls

C) Roughly be able to follow my RL approach demonstrated in https://github.com/backender/ctc-executioner/blob/master/notebooks/analysis_average_price.ipynb e.g. understand that I segmented time and inventory to be able to cancel and resubmit a LIMIT order with the unexecuted inventory at another price level (or submit a MARKET order if time is consumed)

D) Have a look at the order behaviour demonstrated in https://github.com/backender/ctc-executioner/blob/master/notebooks/understanding_events.ipynb Sorry this is not documented yet but just note that

• patterns evolve in the volume map of created or cancelled limit orders
• big trades might accelerate price movement in short term

E) Have a look at the draft of the research objectives (https://github.com/backender/ctc-executioner/wiki#research-objectives) as well as the optimization process with which the questions could be answered (https://github.com/backender/ctc-executioner/wiki/5.-Optimization#process).

[mjuchli]

1. Nothing concrete evolved, except: "How to model data for the purpose of order execution optimization using reinforcement learning".

2. Future progress not discussed due to lack of field knowledge from professor.

3. To be focussed on in future. No intuition regarding an approach provided.

4. Moderate interest.