I recently conducted an interesting experiment with Benjamin, an AI investment assistant, to develop a defensive portfolio strategy in response to escalating Middle East tensions. The prompt was straightforward but specific:
Given the recent Iran and Israel tensions, what portfolio would be best suited for wealth preservation if tensions continue to escalate. The backtest this from Jan 2024 until now and use table where possible to summarize results.
What followed was fascinating (posted in the video above) - Benjamin immediately analyzed recent geopolitical developments, pulling data from multiple sources including Morningstar, Reuters, and Bloomberg. The AI then constructed and backtested a defensive portfolio using five key components: Gold ETF (GLD, 35%), Utilities (XLU, 20%), Consumer Staples (XLP, 20%), Long-term Treasuries (TLT, 15%), and Oil ETF (USO, 10%). This wasn't just random allocation - the portfolio was optimized for minimum risk, achieving an impressive balance of 7.24% annual return with just 7.19% volatility, and a maximum drawdown of only -7.02%.
The most compelling aspect was the AI's comprehensive analysis of each component's performance. Gold emerged as the star performer with a 58.40% cumulative return, while utilities provided stable gains of 28.20%. The portfolio's correlation matrix revealed strong diversification benefits, with most asset pairs showing correlations below 0.30. Benjamin even included specific risk considerations, addressing potential scenarios like oil supply disruptions through the Strait of Hormuz and the impact of interest rates on utilities and treasuries. This level of detail and practical application demonstrates how AI can be leveraged for sophisticated investment strategy development, particularly in responding to complex geopolitical scenarios.
The analysis leveraged modern portfolio theory's mean-variance optimization framework, specifically employing a "MinRisk" objective to construct the most efficient defensive allocation. This mathematical approach helped identify the optimal mix of assets that would minimize portfolio volatility while maintaining reasonable returns, resulting in a portfolio that achieved a strong Sharpe ratio of 1.01 despite the challenging market environment. The optimization considered not just individual asset performance but also their covariances, leading to the specific weightings that provided the best risk-adjusted returns for wealth preservation during geopolitical stress.
Exploring L.A. Wildfire impacts on stocks
Normally this type of analysis would have taken me a few hours at BofA, but with Benjamin AI we can complete it in seconds. The platform automatically pinpoints event dates, pulls price data for the relevant tickers, and runs the cumulative-return calculations—turning what used to be a multi-step manual workflow into a single prompt. You can replicate the same process for hurricanes, payroll surprises, cyber attacks, or any other shock, and even layer on fundamentals (e.g., loss-ratio changes) for deeper insight. Below I ask about the impact of L.A. Wildfires on insurance company stocks.
How do wildfires in Los Angeles impact the stock prices of property and casualty insurance companies? Find relevant date or dates, search for relevant stock tickers impacted by this and then compute cumulative returns from that date of impact to 1 2 and 3 months out and compare to SP 500 as benchmark.
From the video we see that Mercury General Corp was the hardest-hit insurer, falling ≈ -24 % over the three-month window after the L.A. wildfire, versus a -14 % drop in the S&P 500 and roughly flat performance for the other insurers. Mercury endured the steepest decline because of its heavy concentration in the California homeowners market—leaving it far more exposed to wildfire-related claims than its diversified peers.
Tariff example
On March 3, 2025, I began to grow concerned about tariffs and decided to ask Benjamin AI what would make a good tariff-resistant portfolio, as well as to backtest its performance. I sold my positions in Nvidia and other tech stocks and invested in this tariff-resistant portfolio instead. As of May 12, 2025, the recommended portfolio had risen by +2.3%, while the S&P 500 and Nvidia benchmarks had fallen by -9% and -13%, respectively.
The portfolio above I would say is more from a sector perspective but you can also ask Benjamin about first performing stock screens using fundamentals and then backtest this. As an example, I gave the below prompt:
Based on the current conditions in the market with tariffs, create a stock screen for me with twenty names (and create a table) and then backtest this over the last ten years.
It created a screen using real financial data (as opposed to ChatGPT/Groq randomly searching websites) using strong domestic revenue, high profit margins and low debt. It achieved this below backtested results across an equal weight portfolio with twenty names over a ten year period.
📈 Portfolio Annual Return: 15.96% (vs. S&P 500: 11.54%)
💰 Total Return: 408.69% (vs. S&P 500: 167.04%)
⚖️ Risk-Adjusted Performance (Sharpe Ratio): 0.92 (vs. S&P 500: 0.63)
I can then ask Benjamin
what macro data can you use to evaluate the how the tariff situation is going and measure the impact on major indices going as far back as you can
It landed on evaluating import/export prices, trade balance, industrial production and manufacturing indices. To highlight trade balance, it plots the baseline along with 1M and 4M percent changes.
Benjamin also performs a straight forward correlation heat map to ascertain any impacts to major indices but concludes trade balance has low correlation. However, I believe there is more to the story that simple looking at correlations and we will look to include nonlinear ways of measuring causality.
Bitcoin versus Gold
To explore the statistical relationship between Gold and Bitcoin I ask a very simple prompt
what is the statistical link between Gold and BTC going back 10 years
Benjamin explored cross-correlation analysis, performance statistics and rolling correlations.
The cross-correlation study—built from 3,652 daily observations—finds only a very mild synchronised relationship: the peak coefficient is ≈ 0.09 at lag 0, indicating that any link between Bitcoin and gold is both contemporaneous and weak. A handful of other lags (Bitcoin leading gold by ~80–240 trading days) are statistically significant but tiny in size (0.04), suggesting negligible predictive power despite the p-values. In short, neither asset meaningfully leads or follows the other in daily moves.
Then Benjamin extends his thinking to analyze performance summary statistics. Between June 5 2015 and June 5 2025, Bitcoin dwarfed SPDR Gold Trust in performance, posting a roughly 4,062 % cumulative gain and a 42 % average annualized return versus Gold’s 150 % and 7 %, respectively. That spectacular upside came with far heavier risk: Bitcoin’s annualized volatility (57 %) and maximum drawdown (-89 %) were many times Gold’s 12 % volatility and -24 % drawdown, yet Bitcoin still eked out a slightly higher Sharpe ratio (0.74 vs 0.59).
Then Ben turned to measuring rolling correlation between Bitcoin and Gold. For most of 2016-2025 the series hovers close to zero—sometimes slipping slightly negative—signalling that the two assets usually move independently and thus retain diversification value. Only in a few stretches (notably late-2020/21 and mid-2023) does the correlation spike into the +0.3–0.4 zone, and these bursts fade quickly back toward neutrality.
