LP trading bot

[dpaiton]

The LP bot will take some percent of deposits & LP with it; trade with the rest of it.

The LP bot is to:

• explore this as an option to make LPing more attractive
  • we want the LP to make more than the underlying yield source
• learn what the "sweet spot" is for LP vs trading portions your money (is there one?)
• theory is that there will be bigger losses for LPers early on, this makes it less of a blow
  • e.g. fixed-rate drifts up, someone is going to arb that, LP would come out more ahead if they arb it

Configurable params:

• portion of "balance" to LP vs trade
  • the "trade" portion should still sit in yield source if not actively in a position
• define "dead zone" such that if there's a deviation between Fixed/Variable rate that triggers an action
  • make this configurable
  • this is just the arb bot?

Alternative strategies to strictly arbing rates:

• "netting LP bot" -- within any checkpoint, if there is a non-zero net exposure, the LP bot nets it to zero.

Metrics:

• how much money the LP has in their wallet after some amount of time

Experiments:

• Look at performance wrt portion (100% LP vs 100% trade vs in-between)
• Look at how the amount of money matters
  • if the LP can move the market, the strategy should probably change
  • need to test if LP bot is a large vs small % of pool

TODOs:

• [x] add an "lp_and_arb.py" bot

[dpaiton]

Blocked by #919

[dpaiton]

LP BOT STRATEGY

• There exists some parameter sets a "LP factor", l, 0 < l < 1, where the amount you place into more risky trades (longs/shorts) trade_budget = budget * (1-l) and the amount you put into less risky (LP) lp_budget = budget * l

• user sets start budgets by setting a total budget & LP factor.

• what do we do with money won & lost?

  • from the start budgets we can compute proportions, e.g.: lp_constant = lp_budget / total_budget

• primary objective is adding liquidity

  • it is a "liquidity bot" that uses arbitration to make liquidity more attractive

for non-LP amount (= total_budget - lp_budget) we follow the "arb bot strategy"

• sheng's arb bot

  • if fixed_rate > threshold_high: close short; open long for trade_budget
  • if fixed_rate < threshold_low: close long; open short for trade_budget

• better arb bot

  • deadzone is threshold_low < fixed_rate < threshold_high
  • your trade_budget is not in any open positions (should put it into LP position, or variable vault)
  • if (fixed_rate - variable_rate) > open_long_threshold: open long, such that fixed rate is pushed into the deadzone
  • if (fixed_rate - variable_rate) < close_long_threshold: close long completely
  • e.g. close_long_threshold is 0 (fixed_rate < variable_rate)
  • if (fixed_rate - variable_rate) > open_short_threshold: open short, such that fixed rate is pushed into the deadzone
  • if (fixed_rate - variable_rate) < close_short_threshold: close short completely

NOTES

• open_short_threshold will likely be > close_long_threshold
• could set close thresholds to 0 and not expose those to the users

Some alternative bots:

fixed_rate = 5% variable_rate = 2%

open_long == I'm going to earn 5% for 6mo

... time passes ...

situation 1:

current fixed_rate = 3% -- doesn't matter bc I am locked in at 5% current_variable_rate = 10%

close my long, because I could be earning more in the underlying vault

if (fixed_rate_at_open (aka opening_spot_price) < current_variable_rate): close_long

situation 2: current_fixed_rate = 10% -- now I'm losing money current_variable_rate < 10%

I should close my long & re-open a new long if (current_fixed_rate > fixed_rate_at_open) AND (current_fixed_rate > varaible_rate): close_long() & open_new_long()

[dpaiton]

goal for this issue:

• demo a pipeline for spinning up bots with some arb + LP strategy (doesn't need to be perfect)

goal for next issue:

• build an analysis (notebook) pipeline to run through different experiments & plot results
  • we need to be able to objectively compare bot strategies

[dpaiton]

example steps calculate trade size such that fixed_rate will match variable_rate: https://github.com/delvtech/elf-simulations/blob/main/lib/elfpy/elfpy/agents/policies/smart_long.py#L91

[dpaiton]

The elfpy smart long bot will be converted to agent0 in this PR: https://github.com/delvtech/elf-simulations/pull/994

[dpaiton]

resolved by #1000