Only when the tide goes out do you discover who’s been swimming naked.
Warren Buffett
KreamEdge is not a news site and has no ambition to become one. This post is a reaction piece. A story broke this morning that happens to be an unusually clean illustration of something this blog keeps returning to, so it is used here as a case study and then set aside. The news is the occasion. The method is the subject.
On July 30, 2026, the Financial Times reported that Situational Awareness, the AI-focused fund associated with Leopold Aschenbrenner, was seeking additional capital after the sharp selloff in AI stocks. The report said leverage amplified the decline in the portfolio and that the fund had approached investors and lenders for fresh capital.
Two same-day posts on X pushed the story toward the questions a systematic trader would actually ask. A summary from @OdailyChina noted that the fund remained up 439% net through the end of June despite the recent damage. A post from @ZeeContrarian1 focused on the short track record, arguing that one strong wave, even an extraordinary one, does not by itself separate persistent skill from favorable exposure to a market regime.
KreamEdge has not reviewed the fund’s audited return series, prime-broker records, or current book, and holds no view on whether the reporting is complete. Every figure and financing detail above stays attributed to the Financial Times report and the two public summaries. What makes the episode worth writing about is not the fund. It is the analytical failure the coverage exposes: judging risk from the final return number while ignoring the path, the leverage, the concentration, and the financing needed to survive that path.
That failure has a far better teacher than a breaking news cycle. Better System Trader episode 228, recorded in March 2024, put four systematic traders in a room to work through drawdowns, volatility, risk-reward, position sizing, tail exposure, and the bias that appears when current losses make a trader question a long-run process. Andrew Swanscott hosted, alongside Moritz Seibert of Top Traders Unplugged, Jason of Against All Odds Research, and Pavel of Robuxio. Moritz opened with the least marketable sentence in professional trading:
I’m down 99% of the time. The way that I trade, 99% of the time I’m in a drawdown, and then one out of a hundred, if that, you make a new equity high and then you’re going to draw down again. So being a loser, a very professional loser, that’s my job.
Moritz Seibert, Better System Trader episode 228
This article reads today’s news through that discussion. The goal is not to judge one manager or reverse-engineer one portfolio. It is to show how a systematic trader separates volatility from ruin risk, conviction from stubbornness, and evidence from the emotions created by a recent equity peak.
What does the AI-fund drawdown show?
The case shows why a spectacular return does not prove that portfolio risk is controlled. A reported 439% gain through June and a later search for capital after a leveraged drawdown can both be true. The endpoint measures performance from the starting value. It does not measure concentration, peak loss, margin pressure, liquidity, or the ability to keep operating through the next regime.
For a systematic trader, risk management means sizing exposures so the process can survive its expected loss distribution, then defining what evidence would show that the distribution or the process has changed. Volatility is not the same as ruin risk. Leverage, correlation, liquidity, and forced action connect the two.
Key points
- A headline return says little about the leverage, concentration, liquidity, and drawdown required to produce it.
- Volatility can be the source of positive outliers for a convex strategy. It can also become a survival problem when combined with leverage, correlated positions, or unstable financing.
- Win rate is not risk management. A strategy with frequent small wins can still carry a left-tail loss large enough to erase the entire record.
- An index can recover from a drawdown that the individual holding it never recovers from, because one moment of de-sizing at the bottom permanently changes the path that investor actually experiences.
- Drawdowns trigger recency bias, peak anchoring, confirmation bias, action bias, and outcome bias. A precommitted review process helps keep those biases out of sizing and strategy changes.
- Conviction should attach to a tested process with explicit invalidation rules. It should never become a license to ignore a breached risk budget.
- The core objective is survival. A strategy cannot realize its statistical edge if leverage, margin, liquidity, or human tolerance forces it out before the sample develops.
Short records make regime exposure look like skill
The second X post raises a valid inference problem. A strong result observed during one dominant market wave can come from skill, factor exposure, leverage, luck, or a mixture of all four. The return alone cannot identify the mixture.
The claim that a record needs ten years is a useful challenge rather than a universal statistical law. The relevant requirement is exposure to enough independent trades and enough different regimes to test the claimed edge. Four years of concentrated returns can contain less information than a longer, diversified sample, especially when the period is dominated by one theme.
The panel framed the same requirement in terms of sample size rather than calendar time. Moritz Seibert pointed out that a strategy taking one trade per day needs roughly four years just to accumulate a thousand observations, and that this is long enough for most people to quit before the edge has had a chance to show up:
By definition you’ll have one trade per day. To get to a thousand you need four years. So after four years you need to be up. But in that four-year time frame, which is massively long for human beings, every day feels long. You’re very likely to give up.
Moritz Seibert, Better System Trader episode 228
Consider a concentrated AI portfolio during a sustained AI infrastructure boom. Security selection can add value, but the record can also inherit a large common exposure to growth, duration, semiconductor demand, data-center spending, power constraints, and abundant financing. If those positions rise together and then fall together, the ticker count exaggerates the true diversification.
A credible track-record review should therefore ask:
- What was the factor-adjusted return? Compare the portfolio with transparent theme, sector, momentum, volatility, and market benchmarks.
- How concentrated was the economic thesis? Count shared risk drivers, not only securities.
- How much came from changing exposure? Separate security selection from leverage, option convexity, and market timing.
- Did the process survive different regimes? A record should include falling markets, volatility shocks, liquidity contractions, and periods when its main factor is out of favor.
- Was the manager selected after the result? The celebrated survivor is visible. Similar portfolios that failed earlier may not enter the comparison.
- How stable were the decisions? Review the rules and evidence available before each trade, not the explanations written after the outcome.
This is where short-record bias meets outcome bias. Exceptional performance attracts a story about exceptional foresight. A later drawdown attracts the opposite story about obvious recklessness. Both narratives can outrun the evidence. The correct response is to decompose the exposure and test the process across a wider sample.
A 439% return can hide a brutal path
Suppose a portfolio starts at 100 and gains 439%. Its value reaches 539. A later 40% drawdown from that peak leaves 323.4. The portfolio still shows a 223.4% gain from inception, yet the drawdown has destroyed 215.6 units of equity. At a 50% drawdown, the portfolio is still up 169.5%, but half of peak capital is gone.
This is why “still up for the year” is not a risk statistic. It anchors the analysis to the starting value and hides the experience from the high-water mark. It also says nothing about whether the remaining positions can be financed, whether investors can redeem, or whether the strategy can continue at the same size.
Return percentages are path-dependent. A 50% gain followed by a 50% loss does not return capital to its starting point. It leaves the portfolio down 25%. As volatility rises, the gap between arithmetic returns and compounded wealth becomes more important.
A useful risk report therefore needs more than CAGR or year-to-date return. At minimum, inspect:
- Peak-to-trough drawdown and time under water.
- Realized volatility and how quickly it changed.
- Gross and net exposure, including option delta and nonlinear risk where relevant.
- Concentration by position, sector, factor, and thesis. Ten AI-linked positions can still be one trade.
- Liquidity and funding, including margin changes, redemption pressure, and the cost of reducing exposure during stress.
- Expected shortfall and gap scenarios, not only variance measured during calm markets.
The index recovers, the investor often does not
There is a gap between the return an instrument produced and the return anyone actually collected from it. The panel’s clearest illustration was the Nasdaq after the tech bubble. Looked at today, the drawdown is a dip on a long rising chart. Lived through with real money, it was a decision point that most holders failed.
You look at that index now with the benefit of hindsight and go, yeah, okay, every once in a while it goes down, it’s going to be very inconvenient, but look, it’s going up over the long run. What you’re completely forgetting is that very likely you’re going to have given up or altered your bet size. You will either have redeemed, or you will have reduced your investment down there at minus 90%. And what that means is that your experience from that point forward is going to be very different than the ensemble. You can never track the Nasdaq again, because it only takes one occasion for you to not be 100% allocated, and that’s it.
Moritz Seibert, Better System Trader episode 228
This is the mechanism that turns a drawdown into permanent damage without any single catastrophic trade. The loss does not have to wipe out the account. It only has to be deep enough, or fast enough, to force one deviation from the plan. After that, the investor is holding a different position than the one that was backtested, and the recovery belongs to somebody else.
The same logic applies to a leveraged fund facing a margin call. Moritz described the sequence for a strategy that suffers a large but survivable loss:
Even if you don’t lose 100% of your capital, let’s say you lose 60 or 70, it’s almost a probability of one, unless you have cash raining from the skies from other sources, that you will at that point in time change your posture, change the trading system, change the sizing, do something else. And boom, there you have it. You have the deviation from the trading system, and you’re now no longer participating in what could potentially be the recovery.
Moritz Seibert, Better System Trader episode 228
Two consequences follow for portfolio construction. First, the loss limit that matters is not the one at which the account is destroyed, but the one at which the operator, the broker, or the investor base forces a change. Second, diversification across genuinely uncorrelated return streams is not only a return-smoothing preference. It is what keeps the portfolio away from the point where that forced change happens.
Volatility is an engine and a constraint
The Better System Trader panel makes an important point about long-term trend following. Large winners live in the tails of the return distribution. Reaching those tails requires price movement. A strategy designed to cut losing trades and let winners run cannot demand a smooth monthly return at the same time. Drawdowns and giveback of open profit are part of the design.
The stuff that really is the home run that makes you a bunch of money is far out in the tails, stuff that you’ve never seen before. That means volatility, because you can only get to the tail with volatility, and therefore you have to embrace it and not shy away from it. But volatility also means drawdowns. It’s the price that you have to pay. It’s not a penalty, it’s kind of like an inconvenience fee on the way to a bigger bank account.
Moritz Seibert, Better System Trader episode 228
That argument does not mean more volatility is always better. Strategy convexity and portfolio fragility are different things.
- A diversified trend system may accept many small losses to retain exposure to a few large moves.
- A concentrated leveraged portfolio may have several positions that all express the same growth, duration, or liquidity factor.
- A short-volatility strategy may report a high win rate while accumulating an open-ended loss in a rare shock.
All three can look strong in calm periods. Their failure modes are not the same. The first needs enough capital and patience to reach its positive outliers. The second needs concentration and financing limits. The third needs a hard left-tail boundary.
This distinction matters when volatility rises. A market loss can trigger higher margin requirements, wider spreads, thinner books, stronger cross-asset correlations, and investor redemptions at once. The risk estimate made before the move may stop describing the position after the move. Volatility becomes a constraint when it changes the trader’s ability to hold or exit, not only the mark-to-market value.
Win rate is not expectancy
Episode 228 repeatedly separates being right from making money. A strategy can win on only 30% or 40% of its trades and remain profitable if losing trades are bounded while a small number of winners become much larger. The opposite also holds. A 90% win-rate strategy can fail if each winner earns one unit while an occasional loss costs ten or more.
Jason of Against All Odds Research put the trade-off in explicit numbers:
If you’re looking at one-to-one, meaning if I make a dollar, I lose a dollar, I have to be profitable about 60% of the time to make money. 50% obviously breaks even. Then if we jump to the other end of it, if I’m doing one to five, you can be profitable with just 20% of your trades winning.
Jason, Against All Odds Research, Better System Trader episode 228
Expectancy is the relevant starting point:
expectancy = (win probability × average win)
- (loss probability × average loss)Even positive historical expectancy is incomplete. The distribution can change, losses can cluster, and the largest observed loss may not be the largest possible loss. Option selling, martingale sizing, leveraged relative-value trades, and concentrated factor portfolios deserve special scrutiny because their recent win rate can conceal a funding or tail event. The panel was blunt about where a very high win rate usually comes from:
Sometimes you see these strategies where you’re essentially selling tails. You can backtest that stuff, and the backtest will look good. It will have drawdowns, but it doesn’t blow up. Now, the thing is, if you do this long enough, it is not a question of how much will you lose. You will blow up. At some point, you will, statistically speaking, blow up.
Moritz Seibert, Better System Trader episode 228
Pavel of Robuxio made the same point from the strategy-design side, noting that the highest win rates he has built were martingale and short-volatility structures that lose everything in a single session. The failure is not a bad forecast. It is a sizing and tail-exposure choice that the win rate conceals.
A practical position-size calculation starts with the loss budget, not the desired profit:
position weight ≈ portfolio risk budget
÷ estimated adverse move to exitFor example, a 0.5% portfolio risk budget divided by a 10% adverse move implies a 5% position weight before adjustments for gaps, correlation, liquidity, and model error. This is an educational illustration, not a universal sizing rule. If several positions share one factor, their risk budgets must be aggregated. Treating each ticker as independent would understate the portfolio bet.
Bias gets strongest when the risk review matters most
A drawdown does more than reduce capital. It changes the information a trader notices and the actions that feel urgent.
- Recency bias: the latest selloff receives more weight than the full historical sample.
- Peak anchoring: open profit is treated as money that should still be owned, so normal giveback feels like a fresh loss.
- Confirmation bias: the trader searches for data that justifies the decision already preferred, whether that means doubling down or exiting everything.
- Action bias: changing parameters or cutting exposure feels more responsible than waiting, even when the system remains inside its tested range.
- Outcome bias: a profitable trade is labeled good and a losing trade bad without asking whether each followed the approved process.
- Commitment escalation: a public thesis or exceptional prior return makes it harder to admit that assumptions or sizing may need revision.
Peak anchoring deserves particular attention, because it is the bias most directly triggered by a spectacular prior return. Moritz drew a sharp line between the capital originally put at risk and the open profit accumulated on a winning position:
What you’re risking is no longer your core capital. It is no longer the initial risk budget that you carved out. You’re now risking open trade profits, and open trade profits, by definition, in futures markets and derivatives, are a zero-sum game. Somebody else’s money has moved from this account to my account, which to me means I’m now playing with somebody else’s money. And so my thinking is, I have no business really to alter the position.
Moritz Seibert, Better System Trader episode 228
Pavel added the investor-facing version of the same effect, and argued that it is emotionally harder to absorb than an ordinary drawdown, because the profit was already mentally spent:
Probably even worse is giving back open profits. Everyone is always having their Lambos and prepared to keep the money for something. And then immediately, especially on trend strategies, you simply give away 50% of your open profits, 30%, without any problems. This is probably emotionally even worse for clients that are not informed enough.
Pavel, Robuxio, Better System Trader episode 228
The panel also warned that traders often alter a robust system after underperforming peers or suffering a faster drawdown than expected. The dangerous part is that the justification always looks like analysis:
You’re very easily tricked into questioning yourself, questioning the system that you’re currently running when you go into a drawdown, especially if that drawdown happens quickly, if that drawdown is a little bit steeper than your peers. Of course you can find the data to support your thinking at that point in time, but you will be making a decision that is very biased and very strongly influenced by the moment of right now, and it’s not rooted in sample size.
Moritz Seibert, Better System Trader episode 228
His conclusion was that the stamina to leave a tested system alone is itself a source of edge, because so few participants have it. That is the constructive reading. The inverse danger also matters: “stick to the system” can become a story used to ignore evidence that the live process, market structure, or risk budget has changed.
The solution is not blind discipline. It is disciplined falsification.
A drawdown review that separates evidence from emotion
Write the review protocol before the drawdown. Then run the same protocol at fixed thresholds. A useful sequence has four parts.
1. Reconstruct the exposure
Measure current gross, net, factor, sector, volatility, correlation, liquidity, and financing exposure. Recalculate option and leveraged positions under current prices and volatility. Do not rely on the entry-day risk report.
2. Compare the path with the tested distribution
Ask whether drawdown depth, speed, loss clustering, slippage, and correlation fall inside the stress tests. Compare like with like. A 12% drawdown reached in three sessions is not operationally equivalent to the same decline spread across six months.
3. Test assumptions, not recent P&L
Check the conditions that gave the strategy a plausible edge: data integrity, execution rules, liquidity, market access, signal timing, and portfolio construction. A loss does not prove the process is broken. A gain does not prove it is sound.
4. Apply precommitted actions
Define in advance what happens when a portfolio crosses volatility, drawdown, concentration, margin, or liquidity thresholds. Actions can include a staged risk reduction, a pause in new entries, an execution review, or a full strategy hold for investigation. Parameter changes should require a separate research sample and change log. Do not optimize against the drawdown that just caused the review.
Conviction needs an invalidation rule
Systematic trading needs enough conviction to execute through ordinary losses. It also needs enough humility to identify a structural break. These are compatible when the strategy has explicit boundaries.
Asked when trading finally clicked for him, Jason described conviction as something that had migrated away from himself and onto the process:
You will get confident, but it won’t be because you’re overly confident. It will just be more because you understand your systems, you understand your strategies. I’m not confident in myself or my ability. I’m confident in my system and my strategies.
Jason, Against All Odds Research, Better System Trader episode 228
That distinction is what makes conviction auditable. Confidence in a person cannot be falsified. Confidence in a documented process can be, provided the boundaries were written down first. Before deployment, document:
- What drawdown and volatility ranges were observed in sample, out of sample, and in stress tests.
- Which market, liquidity, and execution assumptions must remain true.
- Which events trigger smaller sizing, investigation, or shutdown.
- What evidence is required before changing the model.
- How much capital, margin, and psychological tolerance the full path can consume.
Confidence then belongs to the process, not the last trade, the public narrative, or the size of the prior gain. The purpose of a risk budget is to keep a thesis from becoming an existential bet.
What builders should take from the case
- Report the return path, not only the endpoint. Pair return with drawdown, volatility, concentration, and time under water.
- Aggregate positions by economic factor. A portfolio of different tickers can still be one concentrated thesis.
- Stress financing and liquidity alongside price. A strategy that works only while leverage remains available has a funding dependency.
- Model gap losses and correlation convergence. Calm-period covariance is not a sufficient tail model.
- Size for the drawdown that forces a change in behavior, not only the one that destroys the account.
- Separate process quality from trade outcome. Review whether the approved bet was taken at the approved size.
- Keep a decision journal with the data available at the time. It reduces hindsight edits to both the thesis and the risk plan.
- Precommit the conditions for reducing risk and the evidence required for changing the strategy.
Volatility is not automatically bad. Bias is not automatically eliminated by using code. Both become manageable only when the portfolio has explicit loss limits, realistic stress tests, and enough room to survive the distribution it was built to trade.
Risk management, volatility, and bias FAQ
Is volatility the same as risk?
No. Volatility measures the dispersion of returns. Risk also includes permanent capital loss, leverage, concentration, gap exposure, liquidity, margin, and the chance that a trader cannot continue the process. Volatility becomes a survival threat when it interacts with those constraints.
Can a short investment record prove skill?
No fixed number of years proves skill. The record needs enough independent decisions and enough market regimes to distinguish a repeatable process from factor exposure, leverage, and luck. A concentrated four-year record dominated by one theme can contain less evidence than a longer diversified sample.
Can a high win rate hide serious risk?
Yes. A strategy can win frequently and still lose money if one tail loss exceeds many small gains. At one-to-one risk-reward a trader needs roughly 60% winners to make money, while at one-to-five roughly 20% is enough. Expectancy, loss magnitude, clustering, and the maximum plausible loss matter more than win rate alone.
Why can an index recover when its investors do not?
An index has no funding constraint, no margin call, and no emotions, so it holds its full allocation through every drawdown. An investor who reduces size or redeems near the low permanently changes the path they experience and cannot track the index again. This gap between the ensemble return and the individual return is a core reason to diversify and to size positions for survival.
When should a systematic trader reduce risk?
Risk should change when a precommitted drawdown, volatility, concentration, liquidity, margin, or model-integrity threshold is breached. Reducing risk only because the latest loss feels painful invites recency and action bias.
How can a trader distinguish a normal drawdown from a broken strategy?
Compare the live drawdown’s depth, speed, loss clustering, slippage, and correlations with out-of-sample and stress-test ranges. Then test the assumptions behind the edge: data, execution, liquidity, signal timing, and portfolio construction. A loss alone does not prove failure, and a gain does not prove robustness.
Sources and boundaries
- Financial Times report, July 30, 2026, on Situational Awareness seeking additional capital after the AI-stock selloff. The article is the primary news report cited by both X posts.
- @ZeeContrarian1 post on X, July 30, 2026. The post highlights the reported use of leverage and argues that a short record dominated by one market wave is weak evidence of durable investing skill.
- @OdailyChina post on X, July 30, 2026. The post summarizes the reported losses, capital outreach, leverage, concentration, and 439% net return through June. KreamEdge has not independently verified those figures.
- Better System Trader episode 228, “The Trading Panel, Episode 5,” March 14, 2024, hosted by Andrew Swanscott with Moritz Seibert of Top Traders Unplugged, Jason of Against All Odds Research, and Pavel of Robuxio. The episode discusses drawdowns, volatility, position sizing, risk-reward, investor expectations, and trader bias.
Quotations from episode 228 are transcribed from the published audio and lightly edited for punctuation, filler words, and readability. Meaning and wording are otherwise unchanged, and each speaker is named at the quote. KreamEdge has no affiliation with the Financial Times, either X account, Better System Trader, the panelists, Leopold Aschenbrenner, or Situational Awareness. These citations do not endorse any person, fund, strategy, security, or service.
This article is for informational and educational purposes only. It is not financial advice. Trading and investing involve risk of loss, and past performance does not guarantee future results.
For more research on systematic risk, backtest methodology, and strategy construction, join the free discussion channels on the KreamEdge community page.
0 Comments