WAP support on tickfilter bars

[gnzsnz]

Align ib_async.ticker.Ticker.Bar with TWS API IBApi.Bar class

Reference -> https://interactivebrokers.github.io/tws-api/classIBApi_1_1Bar.html

Change summary

• add wap column
• sort columns aligned with IBApi.Bar.
• set Bar class slots=True
  • using pympler asizeof.asizeof(bar)=736
  • slot version asizeof.asizeof(bar)=152
• include wap calculation on:
  • TimeBars
  • TickBars
  • VolumeBars
• set ib_async.ticker.Ticker dataclasse as slots=True
  • according to pympler asizeof(Ticker) = 7200
  • slotted version asizeof(Ticker) = 3000
• align functionality within Timebars, TickBars and VolumeBars. They all
  emit on new ticks self.bars.updateEvent.emit(self.bars, False) with parameter
  hasNewBar set False. Aligned with reqHistoricalData and reqRealTimeBars

[pavel-slama]

Hi, how about:

new_volume = bar.volume + size
if new_volume != 0:  # Prevent division by zero in empty bar
    bar.wap = ((bar.wap * bar.volume) + (price * size)) / new_volume
bar.volume = new_volume

[pavel-slama]

Seems lika "nan" is the convention for default. If you decide so, also conditions on ln 362 and 398 will need update to isNan(bar.wap).

Although looks like wap can only ever be nan/0 in TimeBar when bar gets created on timer without any input data, so maybe these conditions in Tick and Volume bars are unnecessary?

[gnzsnz]

i'll do a few tests with nan, I just realize that using wap = 0 is running my plots

[gnzsnz]

@pavel-slama please let me know your thoughts.

I have run this a few days with market data and it seems to be working fine.

@mattsta we are using

nan = float("nan") and ib_async.util.isNan, both are available on the standard library. this can be refactored. it requires changes in:

• ib_async.util
• ib_async.ticker
• ib_async.objects
• ib_async.wrapper

probably is too much to push it here, but i can push it into next

[pavel-slama]

@pavel-slama please let me know your thoughts.

I have run this a few days with market data and it seems to be working fine.

IMHO entirely correct solution would keep wap as nan if there's no data - as no price data makes wap undefined not 0. And if the plots should show 0 in such cases - then it should be the plotting code doing the nan -> 0 conversion. But I wouldn't block the merge because of that.

I can imagine it could screw up some code doing trading based on wap getting 0 when in fact there's no data.

(I don't have permissions in this repo anyway, just offering advice)

[gnzsnz]

@pavel-slama please let me know your thoughts.
I have run this a few days with market data and it seems to be working fine.

IMHO entirely correct solution would keep wap as nan if there's no data - as no price data makes wap undefined not 0. And if the plots should show 0 in such cases - then it should be the plotting code doing the nan -> 0 conversion. But I wouldn't block the merge because of that.

I can imagine it could screw up some code doing trading based on wap getting 0 when in fact there's no data.

(I don't have permissions in this repo anyway, just offering advice)

that's what is doing right now. this is from my paper account where i'm running ib_async with this patch.

Bar(time=datetime.datetime(2025, 3, 18, 12, 49), open=20.85, high=20.85, low=20.84, close=20.84, volume=0.0, wap=nan, count=2))

I'm requesting midpoint data, which comes with no volume and wap=nan

[pavel-slama]

@pavel-slama please let me know your thoughts.
I have run this a few days with market data and it seems to be working fine.

IMHO entirely correct solution would keep wap as nan if there's no data - as no price data makes wap undefined not 0.

that's what is doing right now. this is from my paper account where i'm running ib_async with this patch.

Bar(time=datetime.datetime(2025, 3, 18, 12, 49), open=20.85, high=20.85, low=20.84, close=20.84, volume=0.0, wap=nan, count=2))

I'm requesting midpoint data, which comes with no volume and wap=nan

Then it's all good +1

[mattsta]

Interesting improvement!

Some notes:

We do populate wap data in the decoder for BarData in some places (but it's just called average in some places for the object) — though, i'm not sure why it's not in the actual ticker.Bar object (I guess because only RealTimeBar is an API bar with live data?):

ib_async/ib_async/decoder.py

Lines 477 to 511 in e457696

	def historicalData(self, fields):
	_, reqId, startDateStr, endDateStr, numBars, *fields = fields
	get = iter(fields).__next__
	for _ in range(int(numBars)):
	bar = BarData(
	date=get(),
	open=float(get()),
	high=float(get()),
	low=float(get()),
	close=float(get()),
	volume=float(get()),
	average=float(get()),
	barCount=int(get()),
	)
	self.wrapper.historicalData(int(reqId), bar)
	self.wrapper.historicalDataEnd(int(reqId), startDateStr, endDateStr)
	def historicalDataUpdate(self, fields):
	_, reqId, *fields = fields
	get = iter(fields).__next__
	bar = BarData(
	barCount=int(get() or 0),
	date=get(),
	open=float(get() or 0),
	close=float(get() or 0),
	high=float(get() or 0),
	low=float(get() or 0),
	average=float(get() or 0),
	volume=float(get() or 0),
	)
	self.wrapper.historicalDataUpdate(int(reqId), bar)

ib_async/ib_async/wrapper.py

Lines 801 to 819 in e457696

	def realtimeBar(
	self,
	reqId: int,
	time: int,
	open_: float,
	high: float,
	low: float,
	close: float,
	volume: float,
	wap: float,
	count: int,
	):
	dt = datetime.fromtimestamp(time, self.defaultTimezone)
	bar = RealTimeBar(dt, -1, open_, high, low, close, volume, wap, count)
	bars = self.reqId2Subscriber.get(reqId)
	if bars is not None:
	bars.append(bar)
	self.ib.barUpdateEvent.emit(bars, True)
	bars.updateEvent.emit(bars, True)

ib_async/ib_async/decoder.py

Lines 80 to 82 in e457696

	50: self.wrap(
	"realtimeBar", [int, int, float, float, float, float, float, float, int]
	),

For the collective bars, it would take a little more work because the correct vwap calculation is sum(price * size) / sum(size), so we need to actually keep a full history of everything submitted to the collective bar, then when the bar is updated/complete, we calculate the total.

Also the average or wap field could be a @property decorator so it's calculated on-demand from the stored data (the bars list appended to).

So we can either update the wap/average calculation to only happen just before the "complete" time/tick/volume bar is emitted, or make it a dynamic implementation:

@property
def average(self) -> float:
    if self.ticks:
        pv = 0
        v = 0
        for tick in self.ticks:
            pv += tick.price * tick.size
            v += tick.size

        assert v
        return pv / v

    return float("nan")

something like that, but we'd also have to add a new field for tracking all the individual (price, size) tick updates in the collective bars too.

Or (if we have no use for the individual ticks for anything else), it could be done in more of an O(1) fashion by just tracking self.pv += price * size on every update, then before the bar is emitted just populating wap = self.pv / self.volume.

[gnzsnz]

please let me know if more changes are needed to push this PR.

I do acknowledge that setting Ticker to __slots__=True might be pushing too much. I have tested it for many weeks without any problem, but i have not tested L2 data for example. When i measured the gains i told to myself this is too good to leave it for later.