The trap of the one perfect strategy
Retail trading culture runs on the search for a single great system. Tweak the entry, add a filter, rerun the backtest, repeat until the curve looks clean. I spent years in that loop, and it has a structural flaw that no amount of tuning fixes. One strategy is one bet on one market behavior persisting. When that behavior fades, and market behaviors do fade, your entire account rides on how fast you notice and how gracefully you stop.
Hedge funds solved this problem decades ago, and not by finding better single strategies. One brilliant strategy is a bet. Ten proven, uncorrelated, individually mediocre strategies are a business. That idea drives everything below: the strategy portfolio I run live, the process that decides which strategies earn a slot, and an honest look at the first month of live results.
What hedge funds actually diversify
Ask a retail trader about diversification and you get a list of assets: some BTC, some ETH, maybe an index future on the side. Ask a multi-strategy fund and you get a list of return streams. The distinction matters because assets that look different often move together, while two strategies on the same asset can be nearly independent if they profit from different behaviors.
The large quant shops run this at industrial scale. Renaissance reportedly built its results on a huge number of small signals, none impressive in isolation. Multi-strategy platforms allocate across dozens of teams for the same reason: no single desk is trusted to be the whole business. The model is consistent everywhere it works. Many modest edges, combined so their bad periods do not line up.
The math behind it is the same portfolio math that applies to assets. Combining uncorrelated return streams cuts volatility and drawdown without cutting expected return. Two uncorrelated strategies with similar volatility, equally weighted, run at roughly 29% less portfolio volatility than either one alone. I covered why correlation between markets quietly breaks naive diversification in a separate post, and everything there applies one level up, at the strategy layer. The strategy layer is also where you have the most control, because strategy correlation is something you can engineer by choosing what to build next.
The portfolio I run, with honest framing
What I show in the video is the portfolio I run live: a set of automated strategies on crypto and futures markets, each researched, tested out of sample, and promoted through the same pipeline. The combined backtest shows roughly 67% CAGR with about 17% maximum drawdown.
Two caveats before anyone gets excited. First, those are backtest numbers. Backtested performance routinely degrades live, which is exactly why the live-testing section of this post exists. Second, the individual strategies are unremarkable. Taken alone, most are the kind of system a strategy-shopping trader would discard: modest returns, ordinary drawdowns, nothing worth a screenshot. The profile comes from the combination. Their drawdowns do not line up, so one strategy's losing streak lands while another is flat or making new equity highs, and the portfolio curve ends up smoother than any component deserves.
Getting there took infrastructure more than inspiration. An AI-assisted research cockpit manages the pipeline from idea to verdict. A chart lab lets me review individual trades and understand how each strategy actually behaves, not just what its summary statistics claim. It is the same research discipline I described in the order flow research pipeline post, applied to portfolio construction. The denominator matters too: for every strategy that made it into the book, several more died in testing. Most strategy ideas fail out of sample, and a pipeline that does not kill most of its candidates is not testing them hard enough.
How a strategy earns a slot
A pretty backtest is not enough to get a strategy into the book. The candidate has to add something the portfolio does not already have.
The first check is the correlation matrix. I correlate the daily PnL of the candidate against every strategy already running. If it clusters with existing strategies, it is more of the same exposure wearing a new name, and adding it would concentrate risk while pretending to diversify. This check alone filters out a surprising share of candidates, because ideas researched in the same market with the same data tend to rediscover the same underlying bet.
The second check is where most retail analysis stops short. A correlation number computed over the full history is dominated by calm periods, and calm-period correlation is precisely the number that lies to you. Correlations converge under stress. Strategies that look independent through a quiet quarter can draw down together when volatility spikes and liquidity thins. So beyond the overall matrix, I look at tail behavior directly: take each existing strategy's worst days and check what the candidate did on exactly those days. A candidate that stays flat or positive through the book's worst days is worth more than its average correlation suggests. A candidate that quietly loses on those same days is importing tail risk that a single Pearson number will never show.
If you want to run the first check on your own book, the correlation matrix tool we built is free and computes this for crypto assets. The tail check you have to do yourself, and the recipe is short: sort by each return stream's worst days and read the other columns.
First month live: +4%, and what that number is worth
The portfolio's first month live closed at roughly +4%.
I want to be precise about what that means, because this is where trading content usually starts lying. One month proves nothing statistically. A few dozen trades cannot distinguish a real edge from luck, and a +4% month is entirely consistent with a strategy set that goes on to lose money over the next year. Anyone presenting a one-month return as evidence of an edge is selling something.
The reason to go live early, in small size, is different. Live trading is the only place to measure the gap between backtest and reality: actual fills versus assumed fills, realized slippage versus modeled slippage, fees, funding, latency, partial fills. The question the first months answer is whether the live equity curve tracks the backtest minus expected costs. If it does, the model of the world behind the backtest is roughly right, and size can grow slowly. If it diverges, I want to find out with small money, before scaling, while the fix is cheap. I wrote about this discipline in detail in the sim-to-live parity post.
So the honest summary of month one: execution matched expectations, the return sat inside the range the backtest distribution implies, and none of that proves anything yet. Live testing is a required step in the pipeline, not a victory lap.
What a retail trader can copy
None of this requires a fund's budget. The parts worth copying are habits, and they compound.
- →Stop hunting one perfect system. Collect several honest, modest ones instead. A strategy you would discard as boring may be exactly what your portfolio is missing.
- →Test out of sample and let most ideas die. If everything you research passes, your testing is broken, not your luck.
- →Correlate PnL streams before adding anything. A candidate that clusters with your existing book adds risk while posing as diversification.
- →Check worst-day behavior, not just the average correlation. Calm-period numbers understate how strategies move together in stress.
- →Go live small and measure parity. The first months of live trading exist to measure the backtest-to-live gap, not to make money.
The uncomfortable part is the timeline. Building a portfolio this way takes months of research per surviving strategy, and most of the work produces dead ends. That is slower than buying a bot, and it is also the only version of this that has a chance of working.
The standard disclaimer, meant sincerely rather than legally: trading is hard, most strategies fail testing, backtest numbers degrade live, and nothing in this post or the video is investment advice. It is a log of a process, published because the process is the part worth copying.
If you are building your own portfolio of strategies and want a second set of eyes on the research or the infrastructure, that is the work we do. The video above is the full walkthrough.
Frequently asked questions
Why do hedge funds run many strategies instead of one?
Because a single strategy is a concentrated bet on one market behavior persisting, and market behaviors decay. Combining strategies whose returns are uncorrelated lowers portfolio volatility and drawdown without lowering expected return, which is the closest thing to a free improvement that trading offers. It also makes the operation survivable: when one strategy degrades, the others carry the book while it is investigated or retired.
How many trading strategies do you need for diversification to matter?
The benefit starts at two. Two genuinely uncorrelated strategies with similar volatility, equally weighted, cut portfolio volatility by roughly 29% versus running either one alone. Each additional independent strategy helps less than the last, and the real bottleneck is finding return streams that are actually independent, not hitting a count. Five to ten strategies with low pairwise correlation is a realistic target for a serious retail operation. Fifty correlated ones are worth less than three independent ones.
How do you measure correlation between trading strategies?
Correlate the daily PnL streams of the strategies, not the prices of the instruments they trade, using pairwise Pearson correlation over rolling windows. Then go beyond the single number, because correlations measured in calm markets converge under stress. Check tail dependence directly: take each strategy's worst days and read how the others performed on exactly those days. A pair that looks uncorrelated on average but loses together in stress provides far less diversification than its correlation number suggests.
Why do live results differ from backtests, and what can one month of live data tell you?
Live results differ because backtests simplify execution: real fills are worse than assumed, slippage and fees accumulate, some limit orders never fill, and latency shifts entries. One month of live data cannot tell you whether your edge is real, since a few dozen trades cannot separate skill from luck. What one month can tell you is whether execution matches the model. If live PnL tracks the backtest minus expected costs, the backtest assumptions are roughly right and the strategy earns more time and, slowly, more size.
Is your portfolio actually diversified?
Use our free Correlation Matrix tool to compute correlation matrices for crypto assets and see whether your positions are genuinely independent or one bet in disguise. Or book a free 30-minute diagnostic call to talk through your portfolio construction.