Research note
Convexity, Edge, Risk Management
A trader's guide says you need convex bets, a real edge and Kelly-based sizing. We check each idea against Kelly, Thorp, Van Tharp and backtest research.
Short answer: a trader who writes as _dpg argues that individual traders need three things at once: bets that can win much more than they lose (convexity), a real reason to expect profit (edge), and a sizing rule they follow every time (risk management). The sizing part holds up best against the research. The other two are harder to prove than the guide makes them sound.
This post walks through the ideas in _dpg's guide to making money for individual traders, a 14-page PDF, and checks them against published work. The ideas and the examples are _dpg's. The commentary, the outside sources and any mistakes are ours.
What does convexity mean for a trader?
_dpg's definition is simple: a bet is convex "if it can make a lot more money than it can lose" (p. 4). A bought option is the standard example. The most you can lose is the premium, and the upside can be many times that. The guide wants this from every position. It also says plainly that the cost is a low hit rate, with many of these bets winning roughly a quarter of the time (pp. 4–5).
The idea has academic backing in a related form. Nassim Taleb and Raphael Douady define fragility as negative sensitivity to volatility and link it to nonlinear effects. An exposure that is hurt by wider swings is fragile, and one that benefits from them is antifragile. A long option sits on the antifragile side.
Our own exit research shows the trade-off in practice. In MRT's exit-model sweep, a long top-10 basket of post-earnings-beat stocks held for 20 days was tested with a fixed 2×ATR stop. ATR is J. Welles Wilder's volatility measure from his 1978 book New Concepts in Technical Trading Systems. Removing the 2R and 3R profit targets lowered the win rate from 43.1% to 37.4%. CAGR went from 10.2% to 16.1% and max drawdown fell from 14.0% to 12.3%. The panel is current S&P 500 members over 13 years, so survivorship bias inflates the levels, and the study says to read the difference between exits rather than the absolute numbers. In this test, capping winners cost more than the extra wins added back, which fits _dpg's point about not taking small winners (p. 13). These are historical backtest results, not a forecast.
Where does an individual trader's edge come from?
The guide's claim is that a small account can go where large funds can't. Big managers answer to clients and committees, avoid positions that are hard to explain, and leave money on the table when they move size. _dpg's way of finding that money is to follow large buyers and sellers through off-exchange "dark pool" volume, using a third-party data service the guide names.
Two cautions apply. First, this data is public and arrives late. FINRA publishes over-the-counter trading information for each alternative trading system on a delayed basis, so anyone can read it. Whether it predicts price is something you have to test. Second, the guide treats an option's delta as the market's probability of the option finishing in the money. Delta is close to that implied probability but not the same thing, and the implied probability is a risk-neutral one, not a forecast of what will actually happen. Comparing your own odds to delta is a reasonable starting point. It does not prove you have an edge.
If you want to test a flow signal rather than take it on faith, MRT's dark pool backtest page (/dark-pool/backtest) runs an unusual-flow z-score signal through history and reports long and short outcomes against SPY. Our own internal research log records several flow and rotation ideas that did not survive testing, so a flow signal is something to test, not assume.
How does the Kelly criterion size a bet?
This is the part _dpg rates hardest, and it is the part most of the outside research below addresses. The guide says that even a real edge "will not save you from bad position sizing" (p. 7) and points traders to fractional Kelly.
John Kelly's 1956 paper showed how a bettor with an edge can maximize the long-run growth rate of capital. For a bet that either pays off or loses the stake, the Kelly fraction is the probability of winning minus the result of dividing the probability of losing by the net odds. MacLean, Thorp and Ziemba summarize its strengths and weaknesses: Kelly maximizes long-run growth, but its suggested bets can be very large and risky in the short run. Under continuous-time assumptions, betting twice the Kelly amount drives the growth rate down to the risk-free rate. They also warn that the calculation is very sensitive to errors in your estimates, so bets should be cut to be safe.
The guide's worked examples show how much that sensitivity matters. In one hypothetical (p. 9), a cheap put costs $0.12 and the trader thinks it has a 20% chance of being worth $1.50. Kelly says to stake about 13.6% of capital. If the probability is really 10%, the answer drops to about 2.8%. Halving one guess cuts the bet by about 80%. _dpg's response is to distrust personal probabilities for unlikely events, use a small fixed stake per trade (2.5% in the guide), and cap how many positions are open at once.
One technical note on the examples. The Kelly formula uses net odds, meaning the profit on a win per dollar risked. Some of the guide's examples divide the full payout by the cost instead. In the put example the net gain is $1.38 on $0.12, or 11.5 to 1, not 12.5 to 1. That gives about 13.0% at a 20% probability and about 2.2% at 10%. The gap is small, but it always points the same way: gross odds make bets look bigger than they should be.
How do you measure risk and results in R?
Van Tharp's R-multiple is another way to look at the same sizing problem. R is the initial risk on a trade, set by the stop you decide on before you enter. A trade's result is then expressed as a multiple of that risk, and expectancy is the average R across many trades. With convex bets most trades end near minus 1R, so the question is whether the occasional large winner covers them. Win rate alone can't answer that.
MRT has free calculators for this: the R-Multiple Calculator (/tools/r-multiple), the Expectancy Calculator (/tools/expectancy), the Position Size Calculator (/tools/position-size), which works out shares from account size, risk percentage and stop distance, and the Options Payoff Calculator (/tools/options-payoff), which draws expiration P/L for long calls, long puts, covered calls and bull call spreads. The trade journal (/journal) computes each closed trade's R-multiple against the stop recorded at entry. Moving the stop later does not rewrite your R, so expectancy stays comparable across position sizes.
How do you check that an edge isn't overfit?
_dpg tells readers to quantify their edge and run simulations. The danger is that simulations flatter you. Bailey, Borwein, López de Prado and Zhu point out that it is relatively simple to overfit a strategy so it looks good in-sample, and that the more configurations you try, the longer the backtest has to be before a good in-sample result means anything. In their example, five years of data is enough for no more than 45 independent configurations. Try more and a strategy with no skill at all can be expected to show an in-sample Sharpe ratio of 1.
Strategy Lab (/strategy-lab) has three optional switches in its honesty panel that help with this:
- Walk-forward scores the settings you already chose on calendar-year slices of the data, or on two halves when the run covers less than two years. The first slice counts as in-sample, later slices are not used to pick settings, and each slice runs on its own so one year's prices can't change another year's result.
- Trade umpire tags each entry by its opening gap, ATR relative to price, and distance from the 20-day average. Once at least five earlier trades have closed, it blocks an entry whose combination of tags lost on a strict majority of at least three earlier trades, then re-runs the backtest for the kept and blocked entries. It is a fixed rule, not a learned model.
- ATR risk sizes each simulated trade so that a 2×ATR(14) move against it costs about 1% of equity, never more than the run's normal position size. It is a fixed-fraction idea similar in spirit to _dpg's 2.5% rule, applied to stock positions rather than options. It is not Kelly sizing.
None of these can tell you what your odds are on the next trade. They help you see whether a rule you already have survives outside the period you built it on.
What are the limits of this approach?
- The guide's return goals are aspirations, not evidence. _dpg writes openly that the rules were written after a 43% portfolio drawdown, which was later recovered (p. 1).
- Fractional Kelly is only as good as the probability you feed it, and the guide admits that personal estimates for rare events are unreliable.
- A bought option can expire worthless, which means losing the whole premium. FINRA also notes that the leverage in options can come with significant losses.
- Our exit-sweep numbers come from a survivorship-biased panel. They show a direction, not an expected return.
FAQ
Is convexity the same as buying cheap out-of-the-money options?
No. Cheap options are one way to get a convex payoff, but price matters. In the guide's own examples some strikes were rejected because the trader's estimate of their value was barely above the market price. A convex bet that costs more than it is really worth still loses money on average.
What is fractional Kelly?
It means betting a fixed share of the full Kelly amount, such as a half or a quarter. You give up some long-run growth for smaller swings and more room for error in your estimates. MacLean, Thorp and Ziemba describe this as trading growth for security.
Does a low win rate mean a strategy is bad?
Not on its own. Expectancy depends on both how often you win and how big wins are compared with losses. A strategy that wins 35% of the time can have positive expectancy if its average winner is several R. A strategy that wins 70% of the time can lose if its losers are large.
Can MRT tell me my probability of winning a trade?
It can't. MRT's tools measure past results, size trades from a risk number you choose, and check rules on data outside the period they were built on. The probability estimate is still yours, and so is the decision.
Sources
- _dpg, _dpg's guide to making money for individual traders (14-page PDF, undated; supplied to MRT and not publicly linked)
- J. L. Kelly Jr., A New Interpretation of Information Rate, Bell System Technical Journal, July 1956: https://www.princeton.edu/~wbialek/rome/refs/kelly_56.pdf
- MacLean, Thorp and Ziemba, Good and bad properties of the Kelly criterion, January 1, 2010: https://www.stat.berkeley.edu/~aldous/157/Papers/Good_Bad_Kelly.pdf
- Wikipedia, Kelly criterion (formula and definitions), accessed 2026-10-17: https://en.wikipedia.org/wiki/Kelly_criterion
- Taleb and Douady, Mathematical Definition, Mapping, and Detection of (Anti)Fragility, arXiv, August 6, 2012: https://arxiv.org/abs/1208.1189
- Van Tharp Institute, A Short Lesson on R and R-Multiples, date not stated: https://vantharp.com/wp-content/uploads/2018/06/A_Short_Lesson_on_R_and_R-multiple.pdf
- Van Tharp Institute, Tharp Think Trading Concepts, accessed 2026-10-17: https://vantharpinstitute.com/tharp-think-trading-concepts/
- StockCharts ChartSchool, Average True Range (ATR) and Average True Range Percent (ATRP), accessed 2026-10-17: https://chartschool.stockcharts.com/table-of-contents/technical-indicators-and-overlays/technical-indicators/average-true-range-atr-and-average-true-range-percent-atrp
- Bailey, Borwein, López de Prado and Zhu, Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance, Notices of the AMS, May 2014: https://www.ams.org/notices/201405/rnoti-p458.pdf
- Bailey, Borwein, López de Prado and Zhu, The Probability of Backtest Overfitting, revised February 2015: https://www.davidhbailey.com/dhbpapers/backtest-prob.pdf
- Wikipedia, Greeks (finance), delta as a proxy for probability, accessed 2026-10-17: https://en.wikipedia.org/wiki/Greeks_(finance)
- FINRA, OTC (ATS and Non-ATS) Transparency, accessed 2026-10-17: https://www.finra.org/filing-reporting/otc-transparency
- FINRA, Options (investor education), accessed 2026-10-17: https://www.finra.org/investors/investing/investment-products/options
- MacroMomentum Terminal, exit-model sweep and negative-findings log (internal research, not publicly linked)
Educational only, not financial advice. Past and backtested results do not guarantee future returns. Do your own research or speak with a licensed professional before making investment decisions.