Help @mia

Hi mutlaq,

Taking your five questions in order:

1) Filtering by float

The Morningstar dataset on QuantConnect doesn't carry a free-float field — shares_outstanding (on f.company_profile) is total shares outstanding, not tradeable float. The closest available fields are:

• f.company_profile.shares_outstanding — total shares outstanding
• f.company_profile.share_class_level_shares_outstanding — per share class
• f.earning_reports.basic_average_shares.value — basic weighted average shares
• f.financial_statements.balance_sheet.ordinary_shares_number.value and treasury_shares_number.value

Since float ≤ shares outstanding, filtering shares_outstanding < 10_000_000 is a conservative float filter: everything that passes genuinely has a float under 10M, but you'll also drop stocks whose float is under 10M while total shares are above it. For a low-float strategy that's usually an acceptable trade-off.

2) Zero trades / empty stock_data

The usual suspects, in order of likelihood:

• Your filters return an empty set. A combined market-cap ($5M–$100M) + shares-outstanding (<10M) screen is very tight once you've also required has_fundamental_data — many nano-caps have no Morningstar coverage at all and are excluded before your fine filter even runs. Log the counts (see #5) before assuming the trading logic is broken.
• Price is zero when you try to order. Securities added by universe selection have no price until their first bar arrives — at daily resolution that's the next market close. If your scheduled scan fires the same morning a symbol was added, set_holdings will fail silently or warn about zero price. Guard with if not self.securities[symbol].has_data: continue (or check price != 0).
• if self.is_warming_up: return at the top of your event handlers will also block all orders if your warm-up is misconfigured (see #3).

3) RollingWindow never fills at daily resolution

set_warm_up only pumps data through securities that are already subscribed when the algorithm starts — universe-selected symbols join later and get no warm-up at all, so their windows sit empty until 20 live trading days have passed. The fix is to seed the window with a history request when the security enters the universe:

def on_securities_changed(self, changes):
    for security in changes.added_securities:
        symbol = security.symbol
        self.stock_data[symbol] = SymbolData()
        history = self.history(symbol, 20, Resolution.DAILY)
        for bar in history.itertuples():
            self.stock_data[symbol].window.add(bar.close)
    for security in changes.removed_securities:
        self.stock_data.pop(security.symbol, None)

With that in place you don't need set_warm_up for the universe at all.

4) Reliable fields for micro caps

Most reliable, because they come from the price file rather than financial statements: f.price, f.volume, f.dollar_volume, f.market_cap. Next tier: f.company_profile.shares_outstanding and market_cap from the company profile. There is no float field (see #1). Deep financial-statement and ratio fields get increasingly spotty as market cap shrinks — always guard against None/zero before comparing. Also note that requiring has_fundamental_data is what defines coverage: companies Morningstar doesn't cover (a lot of the $5M–$50M space, especially OTC) never reach your fine selection.

5) Counting stocks at each filter stage

Log one summary line per selection pass, with the time included:

def fine_selection(self, fine):
    fine = list(fine)
    by_cap = [f for f in fine if 5e6 <= f.market_cap <= 100e6]
    by_float = [f for f in by_cap if 0 < f.company_profile.shares_outstanding <= 10_000_000]
    self.log(f"{self.time}: fine={len(fine)} cap_filter={len(by_cap)} shares_filter={len(by_float)}")
    return [f.symbol for f in by_float]

Three reasons your Debug() calls vanish:

• Log quota — free tier gets 10 KB per backtest; once it's exhausted everything after is dropped silently.
• Rate limiting — debug() is throttled to ~1 message/second and capped at ~200 characters; excess calls are dropped. Prefer self.log() for diagnostics.
• Duplicate suppression — identical strings are only written once. Always embed a changing value like self.time so each line is unique.

One line per day from universe selection survives all three.

Best regards,
Alex