Held vs Not-Held Orders: Broker Discretion

The difference between held vs not-held orders comes down to one question: is your broker obliged to fill you right now, or do they get room to choose the moment? A held order is an instruction to execute immediately at the prevailing market, and the broker is held to that immediacy. A not-held order hands the broker discretion over time and price, and that discretion removes any obligation to fill at a particular instant. Retail market orders are effectively held. Block and algorithmic orders are usually not-held.

Held vs not-held orders: the core difference

"Held" is short for held to the market. The instruction is plain: take what is there, now. The broker cannot wait for a better print, and cannot leave the tail of the order alone while the quote settles down. Immediacy is the deliverable. The price is whatever the market shows when the order arrives.

"Not held" changes the deliverable. Time and price become the broker's judgment. A not-held order might rest for an hour, or finish in thirty seconds when the liquidity turns up. What the customer gives up is specific: there is no moment at which the broker was supposed to have filled. A fill that lands after the price has moved away is not, on its own, a breach of the instruction.

That is the accountability difference, and it is the reason a held order is more than a fast market order. Speed is the outcome. The obligation is the thing being bought.

What does immediacy cost?

A held buy order crosses the bid-ask spread on arrival. The spread is the gap between the best price a buyer is publicly showing and the best price a seller is publicly showing, and crossing it is the fee for going first. That gap moves through the day. The panel below averages the quoted spread for two large-cap names in basis points of the midpoint, one basis point being one hundredth of one percent, across half-hour blocks of one ordinary session.

SELECT
    formatDateTime(
        toStartOfInterval(toTimeZone(sip_timestamp, 'America/New_York'), INTERVAL 30 MINUTE),
        '%H:%i')                                       AS et_time,
    round(avgIf(toFloat64(ask_price - bid_price)
                / toFloat64((ask_price + bid_price) / 2) * 10000,
                ticker = 'AAPL'), 2)                   AS aapl_spread_bps,
    round(avgIf(toFloat64(ask_price - bid_price)
                / toFloat64((ask_price + bid_price) / 2) * 10000,
                ticker = 'KO'), 2)                     AS ko_spread_bps
FROM global_markets.cache_stocks_quotes
WHERE ticker IN ('AAPL', 'KO')
  AND sip_timestamp >= '2026-06-17 13:00:00'
  AND sip_timestamp <  '2026-06-17 20:30:00'
  AND bid_price > 0
  AND ask_price > bid_price
  AND toFloat64(ask_price - bid_price) / toFloat64(bid_price) < 0.02
GROUP BY et_time
HAVING countIf(ticker = 'AAPL') > 0
   AND countIf(ticker = 'KO') > 0
ORDER BY et_time

At 09:00 ET, ahead of the opening bell, the AAPL quote averaged 6.57 bps wide and KO averaged 21.21 bps. By 12:00 the AAPL figure was 0.89 bps, against 1.42 bps for KO. The last block on the panel, 16:00, printed 6.72 bps for AAPL. A held order pays the spread of the minute it arrives in, and nobody at the broker is permitted to wait for a friendlier one. If a price ceiling matters more than a clock guarantee, that trade is the subject of market order vs limit order.

What is a not-held order?

A not-held order is the same buy or sell with the immediacy requirement lifted. On an institutional ticket it is a flag, often written NH, sitting next to the size and the name. The customer is telling the desk to work it: pick the venues and the pace, and use judgment on price.

Two things follow. The upside is that a worked order can wait for natural liquidity rather than paying the offer on arrival, and can spread its size out instead of pushing the price with all of it at once. The cost is that the customer holds no fill guarantee at any point in time. If the market runs, the order chases it, and the desk has not failed at anything it was asked to do. Discretion cuts whichever way the session goes.

Why a large order cannot be one trade

Discretion only means something when the order is too big to do in one print. The tape is assembled from a great many small pieces and a few large ones, and that size distribution is why working an order exists as a practice at all.

SELECT
    print_size,
    round(100 * prints / sum(prints) OVER (), 2) AS prints_pct,
    round(100 * shares / sum(shares) OVER (), 2) AS shares_pct
FROM
(
    SELECT
        multiIf(size < 100,  'under 100',
                size < 500,  '100 to 499',
                size < 1000, '500 to 999',
                size < 5000, '1,000 to 4,999',
                             '5,000 and up') AS print_size,
        count()                              AS prints,
        sum(size)                            AS shares
    FROM global_markets.stocks_trades
    WHERE ticker = 'AAPL'
      AND sip_timestamp >= '2026-06-17 04:00:00'
      AND sip_timestamp <  '2026-06-18 04:00:00'
      AND size > 0
    GROUP BY print_size
)
ORDER BY multiIf(print_size = 'under 100', 1,
                 print_size = '100 to 499', 2,
                 print_size = '500 to 999', 3,
                 print_size = '1,000 to 4,999', 4,
                 5)

Prints of under 100 shares were 88.99% of every AAPL trade that day and 22.84% of the shares that changed hands. The 5,000 and up bucket runs the other way: 0.02% of the prints, and 48.77% of the shares. A thin sliver of large prints carries a big slice of the volume, while the count of trades is dominated by the smallest bucket, in one of the most liquid stocks listed anywhere.

Set a 400,000 share order against that distribution. No single counterparty is standing there waiting. The parent order becomes hundreds or thousands of child orders, and the decision of when to release each one is exactly the discretion a not-held instruction grants. Some of those children hide their size, which is the job iceberg orders do.

What benchmark is a worked order measured against?

A held order needs no benchmark. It was filled on arrival, so the arrival price is the score. A worked order can only be assessed against a yardstick agreed in advance. The usual two are the arrival midpoint and the session's volume weighted average price. Desks measured against the close work to the closing auction instead, which is where MOC and MOO cutoff times start to matter.

How far apart can arrival and VWAP sit? The trace below measures the distance between AAPL's daily VWAP and that morning's opening print over 9 sessions in June 2026. VWAP is the average price of the day weighted by how many shares traded at each one.

SELECT
    toString(date)                                                AS session_date,
    round(10000 * (toFloat64(max(vwap)) - toFloat64(max(open)))
          / toFloat64(max(open)), 1)                              AS vwap_minus_open_bps
FROM global_markets.stocks_daily_aggs
WHERE ticker = 'AAPL'
  AND date >= '2026-06-15'
  AND date <= '2026-06-26'
  AND open > 0
  AND vwap > 0
GROUP BY date
ORDER BY date

On 2026-06-15 the gap measured 64.3 bps. On 2026-06-26 it measured 266.3 bps. Each of those distances is room a worked order had available, in one direction or the other, and the sign is unknown at the opening bell. The room is also not the same in every name.

SELECT
    ticker,
    round(avg(gap_bps), 1) AS avg_abs_gap_bps,
    round(max(gap_bps), 1) AS widest_gap_bps
FROM
(
    SELECT
        ticker,
        date,
        abs(10000 * (toFloat64(max(vwap)) - toFloat64(max(open)))
            / toFloat64(max(open))) AS gap_bps
    FROM global_markets.stocks_daily_aggs
    WHERE ticker IN ('AAPL', 'MSFT', 'NVDA', 'SPY', 'KO', 'JNJ')
      AND date >= '2026-06-01'
      AND date <= '2026-06-30'
      AND open > 0
      AND vwap > 0
    GROUP BY ticker, date
)
GROUP BY ticker
ORDER BY avg_abs_gap_bps DESC

Across June 2026, MSFT averaged 119.6 bps between the day's VWAP and the open, with a widest single session of 396 bps. At the far end of the panel, SPY averaged 43.6 bps on the same measure. Two identical orders in those two names hand a desk very different amounts of rope.

Where each instruction shows up

• A retail market order is held in practice: it goes out for immediate execution, and the confirmation belongs to the same minute.
• A marketable limit order sits close to held. It caps the price and grants no timing discretion.
• A block handed to a broker's trading desk is typically not-held, with the NH marker on the ticket.
• An algorithmic order is not-held by construction. The schedule is the discretion, written as code.

Time in force answers a different question: how long the order stays alive, which time in force covers in full. Who decides the moment of execution is the held question. In options, working an order is closer to normal than exceptional, and the reasons a resting order sits unfilled are laid out in why options orders don't get filled.

How discretion connects to best execution

A broker handling customer orders owes a duty of reasonable diligence in seeking the best terms reasonably available. A not-held instruction does not switch that duty off. It changes what the duty is measured against, moving from the price on the screen at the instant of arrival to the quality of judgment applied across the working period.

The split shows up in public execution statistics. Standardized market center reports exclude not-held orders from their samples, since those measures assume immediacy was the goal, and a worked order never claimed it. Rule 605 and 606 execution reports cover what those disclosures do include. A worked order's quality lives in the desk's own benchmark reporting, and the customer has to ask for it.

The question worth asking your broker

Two questions settle it. Is my order being worked, or is it going out for immediate execution? And if it is being worked, what benchmark is the fill measured against, and do I get to see that measurement? A desk that works orders has an answer ready. A broker that routes everything for immediate execution has a simpler answer, equally worth knowing. The reader with a 200 share order and the reader with a 200,000 share order want different products, and the words held and not-held are how the industry tells them apart.

FAQ

What does not-held mean on an order?

A not-held order gives the broker discretion over the time and the price of execution. The broker is asked to use judgment across a period rather than fill at a specific instant, and the customer accepts that no single fill was ever guaranteed.

Is a retail market order a held order?

In practice, yes. It is routed for immediate execution against the prevailing market, and the broker carries the immediacy obligation. Most retail platforms never print the words held or not-held, since nearly everything they accept is handled on a held basis.

Does a not-held order get a better price?

Sometimes, never by guarantee. Working an order over time can earn the spread instead of paying it, and can soften the impact of arriving all at once. The same discretion means a fill can land well away from the arrival price when the market moves during the working period.

What is the difference between a not-held order and a limit order?

A limit order controls price and nothing else: it fills at the limit or better, whenever the market gets there. A not-held order controls neither price nor timing from the customer's side. Both sit with the broker, inside whatever mandate was agreed.

Every panel above carries the SQL that produced it. Open one, swap in a ticker you follow, and put the same question to your own names on the Strasmore terminal.