My war thesis started with Venezuela and then moved to Iran. Around October, prediction markets had the odds of a US attack on Venezuela around 40%. Forecasters on Metaculus were more like 70%, and some of the top forecasters I followed were closer to 80%. I trusted the forecasters more than the general market, and even 40% looked high relative to ordinary public discussion.

I should also give a shout-out to the Sentinal Global Risk forecasting group. Their weekly forecasting newsletter was one of the sources I relied on most heavily for my wars thesis.

Leading up to Venezuela, I bought oil companies, oil-related exposure, and VIXY, an ETF linked to short-term VIX futures. I did not make defense companies a major part of the strategy. My thinking was that defense company valuations were more likely to reflect longer-term procurement cycles than short-term news about a specific conflict.

After Venezuela, I moved to anticipating war with Iran. I was worried about picking the wrong vehicle, so I diversified across several things I expected to benefit from a war: oil, oil companies, tanker/shipping exposure, and fertilizer companies. I had used volatility products for the Venezuela trade, but I mostly stopped using them during the Iran war.

The results were uneven. Oil and energy did well overall. Shipping was directionally right in the market, but my own result there was modest. Fertilizer was a small loss. I think now that war made fertilizer more valuable, but it also made fertilizer more expensive to produce, so the net effect on fertilizer stocks was negative.

My approach to trading around wars was to buy well in advance of the event and sell when the war actually broke out. That was mostly right.

I used to be a law nerd. In college I read a lot of court opinions and legal scholarship, and I followed law blogs from lawyers, judges, and professors. My view was that Supreme Court justices were mostly sincere believers in their judicial ideologies, and that public coverage often overstated simple partisan explanations.

That background made me think I had an edge on the tariff case. My view was that the Constitution gives tariff authority to Congress, and that originalism gives a clear answer: the president cannot exercise that power so broadly without Congress. I put the odds of SCOTUS ruling against tariffs around 90%. At the time, prediction-market odds were closer to 50%, so I thought the ruling was not fully priced in.

My plan was to buy a basket of stocks that I expected to benefit if tariffs went away, mostly consumer staples and consumer goods. I bought before oral arguments on November 5, because I expected the arguments to reveal how the justices were thinking. The odds did move from roughly 50% to 70% after oral arguments, but my basket did not immediately move much, which made me feel like the trade had failed.

The data told a different story. My tariff tag finished positive, at +1.7%. My subjective impression in real time was that I had not gotten a positive return from these trades. I only realized the result was positive after analyzing the transaction history.

The US credibility thesis was that Trump-era instability would make the world less confident in the US-led system. If people were less sure the US was predictable, they would look for alternatives to dollar-linked assets. The obvious candidates were commodities, especially gold and silver.

I resisted this at first. When I think of gold and silver, I think of old people getting scammed by ads on Fox News into buying or selling gold. If I bought gold and lost money, I would feel embarrassed. But the more I looked into it, the more reasonable the trade seemed. Someone smart I trust who worked in finance also spoke highly of trading gold, which made me take the idea more seriously. The world relies heavily on the dollar, and the dollar relies on confidence in US institutions. If that confidence weakens, it makes sense for some investors and central banks to diversify into gold.

This became my strongest thesis by tag. The "US credibility" basket finished +4.0%.

Gold and silver did very well for me, but the trade also became volatile. In February, shortly after my second daughter was born, I checked the market at 9:30 AM after being awake most of the night and saw my gold exposure drop sharply at the open. I remember seeing something like a 16% drop and immediately sold everything. No stress, no denial, just the simple realization that I was completely unprepared to deal with that kind of move and still had the opportunity to come out ahead.

I believe in AI. It is already useful to me as a programmer, and I think we are early in a long period where people and companies will keep discovering great new ways to use it.

But believing in AI is not the same thing as believing every AI stock is a good buy. The dot-com bubble happened even though the internet really was transformative. A technology can be real, important, and world-changing while many of the companies attached to it still become overvalued and crash.

My working assumption was that software demand is inelastic. There is a lot of software that people and businesses want but cannot currently afford to build. If AI makes software cheaper to produce, there should be more software, not less. That made me more interested in software companies, small companies, cloud infrastructure, and raw materials for compute than in simply buying the most obvious AI winners.

During this period, I became more skeptical that the current frontier model companies are necessarily the best stock-market expression of AI. Companies like OpenAI and Anthropic sell access to the best models, but open source models keep trailing them by only a few months for many use cases. For high-volume tasks that do not require frontier intelligence, cheaper open models may be good enough.

I also became more cautious about Nvidia and other current AI hardware winners. Their valuation depends heavily on GPUs staying essential to AI, and that may not turn out to be true. Future models may be less dependent on GPU architecture. Nvidia's CUDA ecosystem is a real advantage, though maybe a fragile one, but competitors may improve their software ecosystems, and at some price the economics of non-CUDA systems may matter more than compatibility.

The analogy I kept coming back to was electricity. Since electricity was invented, which grew more: electricity companies, or everyone who learned how to use electricity? I am pretty sure the second. If AI is like that, then the biggest winners may not be the model companies themselves, but the companies that use AI to become more productive.

My tech investments included:

1. Small and mid-sized software companies through XSW.
2. Cloud computing companies through SKYY.
3. The Russell 2000 as a broad bet on smaller companies benefiting from cheaper software.
4. Materials and infrastructure connected to compute, including copper and electricity generation.

The tech tag finished positive, at +1.2%. That was profitable, but the thesis was not as clean as "tech went up." Tech struggled for much of the campaign, and I made money because of timing and trade selection.

I listened to a lot of oil experts because oil was one of my main ways to trade the war thesis. They were useful on details: what happens to an oil well if pumping stops, how hard it is to restart production, what routes can substitute for Hormuz, and how oil logistics actually work.

But they were less useful on the whole trade. Oil expertise does not automatically tell you the probability that Iran will strike Gulf energy infrastructure, how markets will price that risk, or how quickly politicians will change their posture. Several experts seemed confident. But what they knew were narrow facts about their domain, not the full chain of uncertainty.

I saw a similar pattern with tankers. I spoke with a tanker expert who seemed deeply immersed in tanker forums and industry news, but he treated major geopolitical contingencies as afterthoughts even while acknowledging that they were central to the trade. He only became interested in tanker stocks after the war had started and their prices had already exploded. BWET, a tanker shipping ETF, had reportedly risen about 1,300% over the prior year and continued rising into the war. By the time the expert developed conviction about the trade, he was late.

The point is not that the tanker expert was wrong about the industry. His analysis may have been sound, but his focus on tanker-specific information made him less responsive to important information outside the industry. Industry expertise was not enough to make money when it identified the opportunity only after geopolitical events had caused the market to reprice it.

The missing skill is probabilistic translation. Forecasting and prediction-market communities train you to convert information, uncertainty, and your own confidence into quantitative probabilities. An expert can know a ton about a domain and still have a subjective feeling that is poorly calibrated. The trade is almost never about one variable. It is about the joint distribution of politics, logistics, timing, market expectations, and other investors' positioning.

After the Iran war started, my main theses had resolved. I was out of good ideas.

The correct thing to do was probably to stop trading and move into an index fund. But this was only four months into a six-month campaign, and one of my goals was to trade a lot. I felt real internal conflict about that. Stopping early felt like abandoning the experiment, but continuing without a clear edge also felt wrong. I kept looking for trades partly because I wanted the campaign to remain active and partly because I thought even bad trades might teach me something if I was trying my best.

That was expensive.

At the outbreak of the Iran war, I was roughly 22 percentage points ahead of the S&P 500. I finished 7.16 points ahead. Almost all of the lost edge came after my original theses were exhausted.

The experience felt completely different before and after the Iran war. Before the war, I felt like I had deeply thought-through plans and was months ahead of the news. I was mostly alone, thinking about issues almost no one was talking about. After the war started, everyone was talking about those issues, and I was mostly scrambling and chasing headlines.

I put money into regional ETFs like Vietnam, Israel, Mexico, and Latin America, thinking that entire regions were diversified enough to function like broad index exposure. That was wrong. Regional ETFs are not just mini-S&P 500s. They are risk assets that investors often abandon faster during global drawdowns. My regions tag finished negative, at -2.3%.

I also bet against the market after deciding the Iran war was going badly, and would get worse. That worked for a few weeks, but then the market recovered quickly after Trump announced that the war was effectively over. Those trades ended up mixed to negative.

The campaign should have had a rule for what to do when the original edge disappeared. If I have no active thesis, the default should be the index.