🤖
Market Making Bot
This is very much a work in progress. Expect things to change frequently. The Python code shown here is available in the Mango Explorer V3 branch.
Traders buy and sell, but it helps when there are reliable entities for them to trade against. And while an individual trader may buy or sell, they typically aren’t doing both at the same time on the same symbol. In contract, a market maker places both buy and sell orders for the same symbol, producing a valuation of the symbol and saying how much they would be willing to pay for some quantity, and how much they would ask to part with some quantity. They literally make a market by always providing a price at which someone can buy and a price at which someone can sell, and profit by the difference between the buy and sell prices - the "spread."
How market makers know what prices to use, how much inventory to offer and how to manage risk are not topics that will be addressed here as successful market makers develop their own strategies and keep these private.
Let’s look at the mechanics of market making on 🥭 Mango Markets.
Let’s start with a simple example. Here’s an actual marketmaker that will cancel any existing orders, look up the current price on a market, place a BUY order below that price and a SELL order above that price, then pause, then go back to the beginning:
#!/usr/bin/env bash
MARKET=${1:-BTC-PERP}
FIXED_POSITION_SIZE=${2:-0.01}
FIXED_SPREAD=${3:-100}
SLEEP_BETWEEN_ORDER_PLACES=${4:-60}
ORACLE_MARKET=${MARKET//\-PERP/\/USDC}
printf "Running on market %s with position size %f and prices +/- %f from current price\nPress Control+C to stop...\n" $MARKET $FIXED_POSITION_SIZE $FIXED_SPREAD
while :
do
cancel-my-orders --name "WSMM ${MARKET} (cancel)" --market $MARKET --log-level ERROR
CURRENT_PRICE=$(fetch-price --provider serum --symbol $ORACLE_MARKET --log-level ERROR --cluster mainnet-beta | cut -d"'" -f 2 | sed 's/,//')
place-order --name "WSMM ${MARKET} (buy)" --market $MARKET --order-type LIMIT \
--log-level ERROR --side BUY --quantity $FIXED_POSITION_SIZE --price $(echo "$CURRENT_PRICE - $FIXED_SPREAD" | bc)
place-order --name "WSMM ${MARKET} (sell)" --market $MARKET --order-type LIMIT \
--log-level ERROR --side SELL --quantity $FIXED_POSITION_SIZE --price $(echo "$CURRENT_PRICE + $FIXED_SPREAD" | bc)
echo "Last ${MARKET} market-making action: $(date)" > /var/tmp/mango_healthcheck_worlds_simplest_market_maker
sleep $SLEEP_BETWEEN_ORDER_PLACES
done
You can run this and watch it place orders.
For example this will run it on the ETH-PERP market, placing a BUY at the current Serum price minus $10 and a SELL at the current Serum price plus $10, both with a position size of 1 ETH. It will then pause for 30 seconds before cancelling those orders (if they haven’t been filled) and placing fresh orders:
mango-explorer worlds-simplest-market-maker ETH-PERP 1 10 30
That’s not bad for 21 lines of
bash scripting! OK, the price-fetching is a bit contorted, but you can see it’s calling:cancel-my-ordersfetch-priceplace-order(BUY)place-order(SELL)sleep
There are several obvious problems with that approach so let’s see if we can do better.
First of all let’s write it in Python instead of
bash, and let’s put it in an object - SimpleMarketMaker - so that the methods can be overriddden allowing different functionality to be swapped in. Let’s try to be a bit smarter about inventory. And let’s add a check on orders to see if existing orders are OK - even though SOL is cheap there’s no point wasting money cancelling and adding identical orders.try:
# Update current state
price = self.oracle.fetch_price(self.context)
self.logger.info(f"Price is: {price}")
inventory = self.fetch_inventory()
# Calculate what we want the orders to be.
bid, ask = self.calculate_order_prices(price)
buy_quantity, sell_quantity = self.calculate_order_quantities(price, inventory)
current_orders = self.market_operations.load_my_orders()
buy_orders = [order for order in current_orders if order.side == mango.Side.BUY]
if self.orders_require_action(buy_orders, bid, buy_quantity):
self.logger.info("Cancelling BUY orders.")
for order in buy_orders:
self.market_operations.cancel_order(order)
buy_order: mango.Order = mango.Order.from_basic_info(
mango.Side.BUY, bid, buy_quantity, mango.OrderType.POST_ONLY)
self.market_operations.place_order(buy_order)
sell_orders = [order for order in current_orders if order.side == mango.Side.SELL]
if self.orders_require_action(sell_orders, ask, sell_quantity):
self.logger.info("Cancelling SELL orders.")
for order in sell_orders:
self.market_operations.cancel_order(order)
sell_order: mango.Order = mango.Order.from_basic_info(
mango.Side.SELL, ask, sell_quantity, mango.OrderType.POST_ONLY)
self.market_operations.place_order(sell_order)
self.update_health_on_successful_iteration()
except Exception as exception:
self.logger.warning(
f"Pausing and continuing after problem running market-making iteration: {exception} - {traceback.format_exc()}")
# Wait and hope for fills.
self.logger.info(f"Pausing for {self.pause} seconds.")
time.sleep(self.pause.seconds)
It’s following these steps:
- Fetch the current price,
- Fetch the current inventory,
- Calculate the desired price,
- calculate the desired order size,
- Fetch the market maker’s current orders,
- If the desired BUY orders and existing orders don’t match, cancel and replace them,
- If the desired SELL orders and existing orders don’t match, cancel and replace them,
- Pause.
You can see this is similar to the steps in the World’s Simplest Market Maker (above), but it’s a bit more complete. Instead of using a fixed position size, it varies it based on inventory. Instead of blindly cancelling orders, it checks to see if the current orders are what it wants them to be.
It’s worth highlighting the use of a
MarketOperations object in the SimpleMarketMaker. Lines like:self.market_operations.place_order(buy_order)
Show a simple interface to market actions that makes for nice, readable code.
What it hides, though, is that the market maker can work with 3 different market types:
- Serum,
- Mango Spot,
- Mango Perp,
The
market_operations object is loaded based on the desired market, so it doesn’t matter (much) to the market maker if the market is Spot or Serum, it still follows the same steps and the market_operations takes action on the right market using the right instructions.Behind the scenes, a similar variance happens with
MarketInstructions. The actual instructions sent to Solana vary significantly depending on market type, but by having a unified MarketInstructions interface those differences can be largely hidden from market making code. (It’s not perfect but this commonality does help in most situations).This can serve as a kind of a Rosetta Stone for Mango Markets. If you know and understand the instructions sent to Serum to place orders, cancel them, or crank the market, you can look at
SerumMarketInstructions to see how those instructions are implemented in 🥭 Mango Markets Explorer. Then you can compare that file with SpotMarketInstructions to see what bits are different for Spot markets (that require Mango Markets Accounts) and what bits are similar. And then you can explore PerpMarketInstructions to see how those same actions are performed on perp markets. We’ve seen a common structure in the previous market makers, so let’s see if we can provide a nice, common approach for actual market making that allows people to write their own strategies for the interesting bits but that has most of the required code already in place.
The main design ideas behind the design are:
- Every interval, a "pulse" is sent to run the market maker code.
- The market maker is provided with relevant "live" data (like balances) but can fetch any other information it requires.
- The main pluggable component is a "desired orders builder." It looks at the state of balances, market, or other data sources, and it provides a list of BUY and SELL orders it would like to see on the orderbook.
- Another component (also pluggable) compares the desired orders with any existing orders, and decides which orders need to be placed or cancelled.
Live data is provided as a
ModelState parameter to the pulse()method, and it is kept live by a websocket connection that watches for changes in the underlying accounts. That doesn’t matter to the market maker code, it can just assume the ModelState parameter provides up-to-date information on balances, group, prices etc.The
pulse()method is called, say, every 30 seconds (again, it’s configurable). The current version of it looks like this:def pulse(self, context: mango.Context, model_state: ModelState):
try:
payer = mango.CombinableInstructions.from_wallet(self.wallet)
desired_orders = self.desired_orders_builder.build(context, model_state)
existing_orders = self.order_tracker.existing_orders(model_state)
reconciled = self.order_reconciler.reconcile(model_state, existing_orders, desired_orders)
cancellations = mango.CombinableInstructions.empty()
for to_cancel in reconciled.to_cancel:
self.logger.info(f"Cancelling {self.market.symbol} {to_cancel}")
cancel = self.market_instruction_builder.build_cancel_order_instructions(to_cancel)
cancellations += cancel
place_orders = mango.CombinableInstructions.empty()
for to_place in reconciled.to_place:
desired_client_id: int = context.random_client_id()
to_place_with_client_id = to_place.with_client_id(desired_client_id)
self.order_tracker.track(to_place_with_client_id)
self.logger.info(f"Placing {self.market.symbol} {to_place_with_client_id}")
place_order = self.market_instruction_builder.build_place_order_instructions(to_place_with_client_id)
place_orders += place_order
crank = self.market_instruction_builder.build_crank_instructions([])
settle = self.market_instruction_builder.build_settle_instructions()
(payer + cancellations + place_orders + crank + settle).execute(context, on_exception_continue=True)
self.pulse_complete.on_next(datetime.now())
except Exception as exception:
self.logger.error(f"[{context.name}] Market-maker error on pulse: {exception} - {traceback.format_exc()}")
self.pulse_error.on_next(exception)
Again you can see the same steps:
- Build a list of desired orders,
- Get the existing orders,
- Compare them and decide what orders to place.
What’s different here is:
- Desired orders are built using a
DesiredOrdersBuilderobject, and most people will probably want to provide their own version with their own strategy. - Existing orders are tracked, rather than having to be fetched.
- Desired and existing orders are compared using an
OrderReconciler. The default version takes atolerancevalue and if an existing order has the same side (BUY or SELL) and both price and quantity are within thetoleranceof a desired order, the existing order remains on the orderbook and the desired order is ignored. - The code builds a list of instructions, and they’re executed in one step. This is faster, more efficient, and can allow cancels and places to happen in the same transaction. (Instruction size can mean this doesn’t happen though, but the
execute()method takes this into account and uses as many transactions as necessary.)
You can see the different parameters the market maker takes by running:
mango-explorer marketmaker --help
mango-explorer marketmaker --market BTC/USDC --oracle-provider pyth-mainnet-beta --position-size-ratio 0.01
We started by saying what prices to use, how much inventory to offer, and how to manage risk are all great questions that will not be adequately addressed here.
They are up to you.
Last modified 9mo ago