Can a Profitable Trading Bot Handle More Capital? Liquidity, Slippage and Strategy Capacity

A larger account is not necessarily a larger order

Adding cash to an account can leave trading unchanged if the order size stays fixed. Scaling a strategy means changing the capital actually deployed, its number of simultaneous positions, its participation in market flow or its execution horizon. A backtest with a constant percentage position rule silently increases order size as account capital grows. That can change the prices available to the strategy even when its signals are identical.

Our earlier market-versus-limit investigation separated fee savings from missed trades. Here we examine a different question: how much quoted depth a proposed size consumes on entry and exit. We use actual public order-book observations, then apply an explicitly hypothetical execution calculation. Observed depth is useful evidence, but does not by itself establish profitable capacity.

Record the book and its provenance before calculating the result

Before collecting observations, we fixed BTCUSDT spot as the primary instrument, three captures at least 20 seconds apart, and cash budgets of 100, 1,000, 10,000, 100,000, 1 million, 10 million and 100 million USDT. We requested 5,000 price levels on each side from Binance’s documented public market-data endpoint. No credentials, account connection or trading order was used. [1] [2]

The three responses arrived during the UTC request intervals below on 26 October 2026. We retained raw bytes, SHA-256 digests, request metadata and server-time calls before and after each book request. We checked positive quantities, unique sorted prices, an uncrossed best quote and increasing update identifiers. These are separate REST snapshots. lastUpdateId is not a timestamp, and increasing identifiers do not establish a gap-free event stream. Local request intervals and server-clock brackets do not identify the exact exchange time of every quoted order.

All three captures and all seven budgets are reported. This is a short prospective measurement exercise, not a representative sample of liquidity conditions or a holdout profitability test. Repeating the live endpoint later will return a different book; the recorded raw inputs and calculation records are retained internally.

S1 fc2c8e288b0f755f2fe1d16862ad6a5b944a3577b41ac3613b574fc57d2f3e39
S2 f9ed09669d92ad0aa78167dbfc1200fe63bf096d1b686c78a0df8471b30b451d
S3 4031506584a0599b7689312182bad12a63d5786c43c3c1973deca74e57339e58

Define the execution sleeve and reserve the fee before buying

Each budget B is a separate available-cash sleeve, including its entry fee. With an assumed fee f = 0.001 per side, the maximum purchase notional is B/(1+f). Walk asks from lowest to highest price until that amount is spent or captured depth ends. Quantity is the sum taken at each price, and buy VWAP is total purchase notional divided by that quantity. Then walk the same snapshot’s bids from highest to lowest for the acquired units.

This is a continuous-quantity counterfactual, not a submitted venue order. We do not round to actual lot increments or enforce account-specific notional, quantity or order protections. Binance documents those filters separately. The calculation also assumes quoted quantities remain available, no competing orders arrive and our hypothetical transactions do not change later behavior. We model neither a delayed real exit nor order-book replenishment. [4]

Both assumed fees are paid in USDT and charged once. They are not an account-specific fee quote; actual commission assets, discounts and charges differ. The ask/bid walk already includes the quoted spread and price deterioration through depth, so we do not add another spread or slippage charge. Binance’s commission documentation provides the relevant account-mechanics distinction. [5]

entry_notional = sum(ask_price * quantity_taken)
entry_fee = f * entry_notional
unused_cash = B - entry_notional - entry_fee
sale_notional = sum(bid_price * quantity_sold)
exit_fee = f * sale_notional
ending_cash = unused_cash + sale_notional - exit_fee
static_cost_bps = 10000 * (B - ending_cash) / B

The same fee rate produces different total costs at larger sizes

For budgets through 10 million USDT, every acquired unit is covered by the captured bid side in each static calculation. The table reports total same-book round-trip cost, including both assumed fees, in basis points of the available cash budget. One basis point is 0.01%. The 100-million budget is not covered by the captured ask or bid depth; assigning it a fully liquidated cost would invent prices outside our observations.

At 1 million, costs span 21.2165–22.8795 basis points across the three books. At 10 million, they span 39.2707–44.2687. These are size-dependent quoted-depth calculations, not measured market impact from an actual trade. Even the very small budgets have about 19.9813 basis points of static cost because the assumed entry and exit fees dominate their tiny quoted spread.

Follow the 1-million budget through both sides of one book

In S1, a 1,000,000-USDT sleeve reserves 999.000999 USDT for entry commission and spends 999,000.999001 on BTC. Walking 105 ask levels acquires approximately 12.42348330 BTC at a VWAP of 80,412.310721. The midpoint before the hypothetical trade is 80,404.005. The buy VWAP is about 1.0330 basis points above that midpoint.

Selling that same quantity through 43 captured bid levels gives a VWAP of 80,402.348251 and gross proceeds of 998,877.230423. Exit commission is 998.877230, leaving 997,878.353193. With no inventory remaining and no external cash flow, the sleeve’s static cost is 2,121.646807 USDT, or 21.216468 basis points. The calculation reconciles cash and quantity; it does not establish that either transaction could actually have obtained those levels.

If this sleeve belongs to a 5-million-USDT account, the same hypothetical cost is about 4.2433 basis points of total account capital, not 21.2165. The unused 4 million does not receive the strategy’s position return. Keep the sleeve, order and whole-account denominators explicit before comparing a larger account with the smaller bot’s percentage profit.

Ask how much future price improvement would pay for the larger order

A static immediate round trip contains no trading forecast, so a negative result is expected. To connect execution to economics, keep each acquired quantity and bid-side depth unchanged, but multiply every hypothetical future bid by 1+g. This is a declared price-lift sensitivity, not a prediction that a trading signal earns g or that future depth will resemble the captured book. The conditional net result is sale_notional × (1+g) × (1−f) minus the cash used to enter.

The exact break-even lift is cash_used/[sale_notional × (1−f)] − 1. For the S1 1-million sleeve it is 21.2616 basis points, slightly higher than the 21.2165 static cost because the future-price lift has a different denominator. With a 30-basis-point lift, the three 1-million calculations leave +7.0518 to +8.7199 basis points on the sleeve. At 10 million, the same assumed lift leaves −14.4015 to −9.3885. A signal whose economics look adequate at one size can therefore fail a larger-size cost scenario without changing its forecast.

We also declared a 20-basis-point lift before capture. It is insufficient in all three displayed sizes below. Neither sensitivity estimates a true expected edge. The appropriate economic comparison needs the strategy’s own prospective payoffs, actual costs and uncertainty; the live-testing investigation explains that distinction.

When the snapshot ends, the unknown book does not disappear

At the 100-million budget, S1’s captured asks support 30,404,511.278176 USDT of entry cash including the assumed fee, or 30.4045% of the target. The walk acquires 376.50736000 BTC. Its captured bids cover only 204.10621000 BTC, leaving 172.40115000 BTC without a sale price in the known depth. That is a boundary of the dataset, not a report that a real exchange order would partly fill or that deeper orders do not exist. Binance explicitly warns that a finite depth snapshot omits quantities outside its captured levels. [2] [3]

The model retains 69,595,488.721824 USDT of unused entry cash. After the sale covered by known bids, modeled cash is 85,929,861.018185 and residual inventory is still 172.40115000 BTC. Marking that inventory at the original best bid gives an accounting value of 13,861,742.064600, but that price cannot be claimed for a further liquidation after the known bids have already been consumed. We leave the fully liquidated return and price-lift hurdle unspecified rather than treating a mark as a fill.

A tight spread does not reveal the exit or the recovery of depth

S1 and S2 share exactly the same best bid and ask, yet their 1-million static costs differ: 21.2165 versus 22.8795 basis points. At the best bid, displayed quantity falls from 7.42169 BTC to 0.07093 BTC between those observations. The 1-million S1 sale consumes 43 bid levels; S2 consumes 197. Looking only at the spread or the entry ask would miss this change in the opposite side.

The three captures cannot tell us how rapidly consumed depth replenishes, how competing bots react or what liquidity survives a stressed exit. The ECB distinguishes depth, tightness and resilience and notes that narrow spreads can conceal little quantity at the best quotes. Original order-book research also studies price changes using event flow, not merely a static spread. Those observations motivate the questions; they do not calibrate an impact or resilience model for these captures. [6] [7]

Treat the next allocation as a new execution and risk test

Define a proposed order size and completion deadline, then evaluate eligible signals over varied conditions with sequencing, latency and account records. Compare buy and sell VWAP with a contemporaneous decision reference and count missed, partial and rejected executions. Account for the current position and other agents’ simultaneous orders; two bots cannot both claim the same displayed liquidity as if each were alone. A passive or sliced policy may change cost, delay and missed-opportunity risk; it requires a separate comparison, not a free reduction of the static walk.

Reserve fees and residual risk capital before opening, and specify the exit plan and scaling rollback. A larger account can retain cash rather than force the same percentage allocation into the market. Fixed operating costs, financing where applicable and market or concentration risk still matter. More capital does not create a larger predictive advantage, and a profitable small pilot does not measure the next size’s execution.

ChatGPT or Claude can help turn a scaling proposal into a test specification; Codex or Claude Code can help audit book traversal and cash identities. Here AI assistance supported the analysis and Python performed decimal calculations on the recorded public responses. No AI trading model, account or actual order was evaluated. Require the report to separate observed book fields, calculated counterfactuals and unmeasured future behavior.

Given these timestamped public books and a declared cash budget, reserve the stated fee, walk asks and bids without inventing deeper liquidity, and reconcile cash and inventory. Report VWAP, consumed levels, known-depth coverage and unresolved inventory. Keep observed quotes separate from hypothetical fills. State which latency, impact, lot filters, replenishment, future returns and account risks are not measured. Do not label a snapshot size threshold a strategy-capacity limit.

What the measured books justify—and what they leave open

The recorded books support a concrete result: under one continuous, static-depth calculation and fixed fee assumption, the modeled implementation burden rises materially between 1 million and 10 million USDT. A declared 30-basis-point future bid lift covers the former scenario but not the latter. At 100 million, the captured data do not cover the proposed round trip at all. None of those facts identifies a universal capital ceiling or proves that the smaller scenario earns money.

Strategy capacity is the allocation for which its whole trading process can still meet its economic and risk requirements. Establishing it needs the return opportunity, execution policy, concurrency, holding horizon and adverse exit conditions together. The next justified step is a bounded size experiment with retained evidence, rather than multiplying a small bot’s historical percentage return by a larger deposit.