Brian Gelber: How Trading Changes with the Marketplace

A hypothetical trading model earns 0.08 per share before costs. Its execution assumptions allow 0.03 per share for the complete trade, leaving 0.05. If the actual cost rises to 0.09, the same signals produce a negative average. Brian Gelber’s career provides a useful setting for thinking about what survives when the way a market trades changes.

Gelber Group’s own history says Brian and Frank Gelber founded a Chicago Board of Trade brokerage in 1982. It describes a later move toward screen-based proprietary trading and an evolution into algorithmic trading. That is evidence of a changing business and execution environment. It doesn’t supply a universal entry formula that a private trader can copy.

The examples below are independent and hypothetical. They examine how information, fills and costs affect a trading process. They aren’t Gelber’s trades, his firm’s proprietary strategies or claims about its returns. The practical lesson is to identify which parts of a method depend on the environment in which it developed.

Brian Gelber and the question of what actually transfers

A historical trading story can contain several different ingredients: access to information, interpretation of that information, the ability to execute and control of the resulting exposure. A modern chart may preserve only some of them.

Consider a fictional floor trader observing an unusually urgent sequence of orders at 100.00. A later reader sees only a daily high of 101.00, a low of 98.00 and a close of 100.50. That daily bar cannot reconstruct the order sequence, its timing or the trader’s opportunity to transact.

I find the difference between the order sequence and the 100.50 close important because both can be called “price action.” They contain different information. A method that relies on one can’t be reproduced simply by examining the other.

The order-flow imbalance guide introduces a more specific vocabulary for activity near the bid and offer. Even with such data, a private trader still needs to understand coverage, latency and how the measurements are constructed before treating them as equivalent to another participant’s information.

A visible quote does not promise the visible fill

Suppose a hypothetical market displays an offer of 25.10 for 200 shares. A model assumes that an order for 1,000 shares can execute entirely at 25.10. That assumption extends beyond the quantity actually shown at that price.

If 200 shares execute at 25.10, 300 at 25.11 and 500 at 25.13, the weighted average is 25.118. The extra 0.018 per share costs 18 on the 1,000-share order. A strategy with a small expected move can be sensitive to that difference.

There may be additional liquidity or changes before the order arrives. An executable model therefore needs assumptions about quantity and timing alongside the recorded best price. The example leaves the eventual fill uncertain.

A limit order introduces a different uncertainty. It constrains the acceptable price, but it may remain unfilled while the market moves away. Counting every touched limit as an automatic fill can exaggerate a simulation, particularly when the record says nothing about queue position or available quantity.

Confirmation changes the entry and the remaining distance

A fictional breakout strategy considers resistance at 50.00. One version enters after a close at 50.20. Another waits for a second close at 51.00. If both use an exit reference at 49.50, the initial distances are 0.70 and 1.50.

With a planned price-risk allowance of 210, the corresponding quantities are 300 shares and 140 shares. The later entry uses more observed evidence and a larger distance per share. It doesn’t automatically justify a larger position.

I would keep the 0.70 and 1.50 distances beside the confirmation rule when comparing the versions. A later signal can avoid some initial failures while also reducing the remaining opportunity or producing different losses. The record has to show the combined result.

The guide to volume confirmation on breakout candles describes another possible filter. Extra volume doesn’t establish that the remaining move will be profitable, and it doesn’t identify the motives of the participants. A filter needs its own comparison with an otherwise consistent baseline.

Costs can erase an apparently repeatable advantage

Return to the opening model with a gross average of 0.08 per share. At a complete-trade cost of 0.03, its net average is 0.05. At 0.09, the net average becomes minus 0.01. This calculation holds the gross results constant to isolate the effect of costs.

In a real investigation, the gross result may change too. Slower execution can alter which trades fill, while larger orders can interact differently with available liquidity. A fixed deduction is a useful first approximation, but it may miss those relationships.

A practical simulation can therefore report several cost assumptions alongside the gross record. If profitability disappears with a change from 0.03 to 0.04, the result is more sensitive than one that remains positive at 0.10. Neither sensitivity test establishes future profitability.

The exercise also discourages selecting an attractive historical return first and adding convenient cost assumptions afterward. The assumptions belong in the research specification. A method that only works at an unachievable fill deserves a different assessment from one with room for execution error.

A changed market needs a dated comparison

Suppose a fictional model makes 200 trades in one period and 200 in a later period. The later record shows lower average gains and more missed fills. Combining all 400 trades into one number can hide when the deterioration appeared.

A review might compare spreads, volatility, traded volume, order size and time of day across the two periods. It could ask whether the signal’s subsequent price movement weakened or whether execution captured less of an unchanged movement. Those explanations imply different research tasks.

Historical volatility provides one description of changing price variability. It doesn’t explain every change in tradability. Two periods with similar volatility can have different spreads, available quantities and responses to orders.

The dates matter because a later rule change can contaminate the comparison. If the model was revised midway through the second period, the review needs to separate the old and new versions. Otherwise it may attribute a design change to the market or blame the design for an execution problem.

Automation makes unclear instructions more visible

“Enter when the breakout looks convincing” cannot be executed consistently by code without further definitions. A hypothetical specification could require a completed close above the previous 20-session high, using only the preceding sessions in the calculation.

Including the current session’s high would change that threshold and could make the close condition impossible or misleading. The lookback convention is a small implementation detail with a large effect on which trades appear.

The Ed Seykota profile offers a related discussion of systematic rules. A precise rule can be inspected and repeated, but precision alone gives it no economic advantage. A fully specified model can lose consistently.

Automation also creates operational questions that discretionary descriptions often omit. If a data update is late, does the model wait, skip the signal or use the last known value? If an order is rejected, what state does the system record? A result that assumes every message succeeds leaves these cases outside its evidence.

Firm history is context, not a private trading manual

Gelber Group’s public description presents a privately funded trading firm operating across asset classes. That description gives context for the business. It doesn’t reveal the position limits, data arrangements or individual models behind its activity.

For a hypothetical private trader, an adaptation exercise could list every required input for one proposed setup: completed prices, volume, timestamp convention and order type. Any item missing from the available data is a limitation to resolve before claiming the historical approach has been replicated.

A daily chart study, for example, may support a next-session signal test. It cannot establish whether an intraday strategy would have filled before an abrupt reversal at 10:03. The evidence needs the resolution required by the decision.

The documented change from a Chicago Board of Trade brokerage to screen-based and algorithmic trading raises a practical question: when the environment changes, which observations and operations still exist? The answer has to come from the current process, rather than from confidence in a famous name.

Read the interview and keep the authorship straight

Market Wizards is Jack D. Schwager’s interview collection. The publisher’s contents for the updated edition lists the Brian Gelber interview beginning on page 345. A useful reading note separates Gelber’s description from the reader’s proposed rule, such as the two-close test above.

Pit Bull, by Martin Schwartz, is related reading about another trader. It doesn’t establish Gelber’s method or serve as evidence for his performance. Comparing the two accounts can be useful if the notes keep each person’s decisions and circumstances distinct.

Disclosure: As an Amazon Associate, I earn from qualifying purchases.

The linked Market Wizards edition is the updated paperback by Jack D. Schwager: (US) Market Wizards, Updated, (UK) Market Wizards, Updated, (DE) Market Wizards, Updated.

Pit Bull is linked in paperback: (US) Pit Bull, (UK) Pit Bull, (DE) Pit Bull.

Keep the fill beside the signal

A model earning 0.08 before costs and losing 0.01 afterward has answered an important question. Its signal and execution assumptions don’t yet support the claimed result. Retaining both numbers makes the next research task concrete, even when the chart still looks persuasive.

Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.