All source code and data handling steps are publicly accessible at the following link:
Google Colab Notebook:
https://colab.research.google.com/drive/15dBsIpnSFBJzyruTeddNcWyrrCyVENL9?usp=sharing
I. Core Strategy: Bet the Dip (Buying When Sentiment Is at Its Worst)
Looking back at 2025, nearly all of my meaningful entries can be summarized under a single core idea: Bet the Dip.
Rather than chasing upward momentum, I chose to enter positions when market sentiment was clearly pessimistic and prices experienced temporary pullbacks.
In practice, most of my positions were established during a major correction in April, a smaller pullback in September, and a broader sell-off across AI-related assets in November.
These periods shared a common pattern. Short-term news flow was consistently negative, and overall market sentiment was cautious. However, from a medium- to long-term perspective, I did not believe the core AI investment thesis had fundamentally broken. That was why I felt more comfortable buying against prevailing sentiment instead of waiting for conditions to “feel safe” again.
II. Overall Results: Returns Driven by Entry Decisions
From a results standpoint, as of year-end:
- Total Realized PnL: $2,171
- Win Rate: 77.78%
- Expectancy (per trade): +$60.31
- Average Holding Period: 36.7 days These numbers are not extraordinary on their own, but they reflect something important:
this year’s returns were not driven by a single lucky trade, but by repeated decisions and relatively consistent execution.
III. Trade Structure: Win-Rate Driven, Not High-Payout
From a structural perspective, my trading style this year was fairly clear:
- Profit Factor: 2.69
- Payoff Ratio: 0.77 In other words, this was not a strategy that relied on a small number of outsized winners to offset frequent losses.
More often than not, gains were small to moderate, while mistakes—when they occurred—tended to be more concentrated in magnitude.
This structure is not inherently flawed, but it places high demands on two things:
position sizing, and avoiding extreme errors.
IV. Best-Executed Trades: Giving the Trend Time
The trades I felt best about this year were primarily concentrated in Rocket Lab (RKLB) and NVIDIA (NVDA).
RKLB is a representative example. After building the position near relative lows, I did not rush to exit. By allowing sufficient time for the thesis to play out, individual trades achieved returns of approximately 95%–99%, with holding periods of around 100 days.
NVDA followed a similar pattern. While I did not capture the entire move, maintaining patience within the trend resulted in a return of roughly 37%.
These trades reinforced one idea for me:
When the direction is right, time itself becomes a source of return.
V. The Most Important Mistake: A Speculative Trade That Should Not Have Happened
If there is one trade that deserves the most reflection this year, it is the GME put option.
It was not a complicated trade to explain—
at its core, it was a short-term, high-risk speculative attempt.
The outcome was straightforward:
- Single-trade loss: approximately $584
- Return: close to 100% The issue was not directional judgment, but a complete failure in position sizing.
Within a trading framework that is largely win-rate driven, an extreme loss like this carries disproportionate destructive power. This trade delivered the clearest and most unambiguous lesson of the year.
VI. Other Losing Trades: Contained, but Worth Reviewing
Beyond GME, there were other losing trades, including:
- CRCL: approximately 18.7%
- AAPL: approximately 10.8% These losses fall within what I would consider a normal range of trading error and were not catastrophic. Still, they serve as a reminder that position size should remain conservative when trends are not fully established.
VII. Long-Term View on AI
To date, my view that AI is better approached with a medium- to long-term investment horizon has not changed.
In the short term, large-scale consumer (2C) applications remain limited. However, in enterprise (2B) use cases—particularly in compute, infrastructure, and enterprise software—the potential for AI adoption remains substantial.
Given my professional background in AI application development, I am inclined to believe that more meaningful and systemic commercialization is likely to emerge from 2026 onward. This belief underpins my willingness to tolerate volatility and deploy medium-sized positions during periods of market pullback.
VIII. Annual Summary: The Real Issue Was Not Judgment, but Tool Mismatch
This was my second year investing in the market. Looking back at 2025, my biggest takeaway was not tied to any single trade outcome, but to a clearer realization of a structural issue:
The primary problem this year was not incorrect directional judgment, but using mismatched risk tools in otherwise correct trades.
In AI-related assets, I was willing to buy during pessimistic sentiment and benefited when trends unfolded. This suggests that direction and timing were not the core issues.
However, at the execution level, I applied short-term profit-taking rules to positions that arguably deserved long-term holding, while simultaneously taking on excessive risk in speculative trades that did not justify it.
The result was clear:
Good decisions were not amplified enough
Bad decisions were amplified disproportionately This led me to realize that the next stage of improvement is not about what to look at, but rather:
Which tools to use to express a view
How much capital to risk when wrong
Which convictions deserve time, and which only deserve a limited attempt If the first year was about learning the market, and the second year was about validating a strategy, then this year was about clarifying a deeper question:
Which returns are truly repeatable—and which are merely accidental.
IX. Plan for the Coming Year: From Judgment to Risk Architecture
Looking ahead to the new year, my focus is not on fundamentally changing what I invest in, but on more deliberately improving how risk is expressed and managed around my convictions.
Based on the lessons from 2025, my plan for the coming year can be summarized in three directions:
1. Clear Separation Between Core Positions and Speculative Trades
Going forward, I will draw a much clearer distinction between:
- Core positions built on medium- to long-term convictions (e.g., AI infrastructure and enterprise applications), and
- Speculative or tactical trades, which are inherently short-term and opportunity-driven. For speculative trades, I will enforce explicit position caps, ensuring that even a total loss cannot materially impact overall portfolio performance.
2. Aligning Risk Tools With Investment Horizon
One of the key issues this year was applying short-term profit-taking and risk controls to positions that were fundamentally intended for medium- to long-term holding.
In the coming year:
- High-conviction, long-term positions will be managed with wider risk bands and longer time horizons
- Short-term trades will continue to use stricter risk controls, but with smaller position sizes The core objective is to ensure that risk management aligns with the underlying investment thesis, rather than forcing all trades into a single framework.
3. Fewer Decisions, Better Amplification of Correct Judgments
Rather than increasing trading frequency, my goal is to:
- Make fewer but higher-quality decisions
- Allow strong convictions to compound over time
- Reduce situations where correct judgments fail to meaningfully translate into results
Closing Thought
If the first year was about learning the market, and the second year was about validating a strategy and understanding myself, then the coming year will focus more on structure—building a risk architecture that allows correct ideas to fully express themselves while keeping mistakes contained.