Portfolio construction gets presented as something mystical — a mix of art and quantitative magic that takes years to master. The reality is more tractable than that. There are a handful of principles that actually drive diversification outcomes, and once you understand them, building a well-structured portfolio from zero is less intimidating.
This guide covers the core logic. It applies whether you're building your first competition entry on Finology or thinking about how to structure a real investment account.
Start With the Question Nobody Asks First
Before picking any stocks or allocations, ask: what's the portfolio for? The answer changes everything about how you should build it.
A competition portfolio with a 4-week window needs to be oriented toward generating alpha over that short period. A long-term personal portfolio should be optimized for risk-adjusted returns over a multi-year horizon, where short-term volatility matters less than avoiding permanent capital loss. These are genuinely different objectives and they produce different portfolios.
The mistake most first-time traders make is building a competition portfolio like they're managing a retirement account — conservative, diversified, designed not to lose — and then wondering why they rank in the middle of the leaderboard. Or building a real portfolio like they're in a competition — concentrated, aggressive, fast-moving — and then experiencing the kind of drawdown that takes years to recover from.
Define the objective first. Build for that objective.
The Mechanics of Diversification
Diversification reduces portfolio volatility when the assets in the portfolio are not perfectly correlated. That's the whole math of it. The less correlated your positions are, the more positions you can add before the marginal diversification benefit drops to near zero.
In practice:
- Two perfectly correlated assets (correlation = 1.0) provide zero diversification benefit regardless of position sizing
- Two perfectly uncorrelated assets (correlation = 0) reduce portfolio volatility to 70.7% of a single-asset portfolio at equal weights
- Two negatively correlated assets (correlation = -1.0) can theoretically reduce volatility to zero at the right allocation
Real portfolios never have perfectly uncorrelated assets — correlations cluster and change over time. But the principle holds: you want positions that tend to respond differently to market events.
Building in Layers
A practical way to think about portfolio construction is in layers:
Layer 1: Market/Beta Exposure
Your portfolio will have some correlation to the overall market regardless of what you pick. Decide how much market exposure you want. A fully hedged portfolio aims for zero correlation to the market (beta = 0). A full long portfolio has high market correlation (beta close to 1). Most competition portfolios are partially hedged — long positions in high-conviction stocks, some offsetting positions to reduce pure market risk.
Layer 2: Sector Weights
Within your overall market exposure, decide how you want to weight across sectors. Equal-weighting sectors is one approach; tilting toward sectors you expect to outperform is another. The key is to do this deliberately — sector tilts are often the largest source of performance difference between competition portfolios, and they're often made accidentally rather than by design.
Layer 3: Individual Position Selection
Within each sector allocation, select the specific stocks. At this stage, focus on: which companies within the sector have the best fundamental drivers right now? Which have the most attractive risk/reward given current valuations? Which are least correlated with each other within the sector?
Layer 4: Position Sizing
How much of the portfolio to allocate to each position. Volatility-adjusted sizing is the most principled approach — allocate more to low-volatility positions and less to high-volatility ones so that each position contributes roughly equal risk to the portfolio. This is counterintuitive when you have a high-conviction idea in a volatile stock, but the math supports it consistently.
The Correlation Trap in Practice
Let's make the correlation trap concrete. Suppose you build a portfolio with these five positions:
| Stock | Sector (Label) | Actual Driver |
|---|---|---|
| Samsung Electronics | Technology | Memory semiconductor cycle |
| SK Hynix | Technology | Memory semiconductor cycle |
| LG Energy Solution | Industrials | EV battery demand + Lithium prices |
| Hyundai Motor | Consumer Discretionary | EV vehicle sales + USD/KRW |
| POSCO Holdings | Materials | Global steel demand + EV supply chain |
On paper, five different sectors. In reality: two positions directly in semiconductors, three positions correlated to global manufacturing and EV demand. If Chinese EV demand slows and global manufacturing softens simultaneously — which has happened — this "diversified" portfolio behaves like a concentrated industrial/tech bet.
Practical Numbers for Competition Portfolios
Based on what we see in competition data on Finology, portfolios with these characteristics tend to perform well on risk-adjusted metrics:
- 8–15 positions: Enough for meaningful diversification, few enough to manage actively
- No single position > 15% of portfolio: Limits single-stock blowup risk
- No more than 3 positions sharing a primary risk driver: Limits hidden correlation concentration
- At least 2 sectors with low correlation to your primary theme: Provides a buffer when your main thesis is wrong
These aren't hard rules — they're starting points that can be adjusted based on your strategy and the competition window length. A 1-week sprint competition can tolerate more concentration than a 3-month championship, because there's less time for diversification to add value.
"The best portfolio is not the one that performs best when you're right. It's the one that doesn't destroy you when you're wrong — while still letting you participate when you're right."
That balance is what portfolio construction is really about. It takes practice to calibrate, and trading competitions are genuinely the best low-stakes environment to develop that calibration before it matters.
