Personal

A lot of policy has hit the economy since January 2025. Tariffs went back on imports from China, Mexico, Canada, and others, which raised costs for manufacturers and for anyone buying what they make. The "Big Beautiful Budget Bill" extended the earlier tax cuts and pushed new money toward infrastructure and defense. Stricter immigration rules and new limits on Medicaid and food stamps came with it.

Paying for all that falls to Treasury Secretary Scott Bessent, who's been borrowing heavily on the short end. The bet is that rates drop and refinancing gets cheaper. So far the Fed hasn't played along. Trump wants deep cuts, Powell won't move, and his reasoning is that inflation hasn't settled and nobody knows yet how the spending and the tariffs feed through.

Looking at the data, wage growth has cooled off, so most workers aren’t seeing much growth in their paychecks. The prime-age employment rate—basically, how many 25- to 54-year-olds are working—has flattened out, which points to a job market that’s losing steam. Credit spreads—the gap between what companies and the government pay to borrow—have actually narrowed a bit, so markets aren’t panicking yet. But the yield curve is still inverted: when short-term rates are higher than long-term, it’s a sign that investors are bracing for rough times. Debt service ratios—share of household income goes toward paying off debt—are creeping up again after a drop during the pandemic. That means more families are shelling out for debt(loans, credit cards, etc...) which could become a problem if rates don’t fall or unemployment rises. On top of that, Google searches for “recession” have been bouncing around, telling you that people aren’t exactly feeling secure.

Meanwhile, the national debt just blew past $36 trillion. Some say heavy borrowing and spending are the price of investing in America, while others warn we’re gambling with the country’s future if interest rates spike or the economy takes another hit. So, are these signals pointing to a real recession on the horizon, or is this just what “normal” looks like in today’s economy? I’ll break down each indicator—jobs, wages, credit spreads, the yield curve, debt, and sentiment—and try to make sense of what’s actually happening. 

## Establishing a Baseline

Before building my own model, I started by recreating the Fed’s basic recession probability model as a benchmark. The Fed’s version is built around the yield curve—the difference between the 10-year and 3-month Treasury rates. When the curve goes negative, it’s often taken as a recession warning.

In my version, I used monthly data (for a closer look at changes as they happen) and ran the numbers using Python with up-to-date FRED data. My setup and logic matched the Fed’s as closely as possible, but because I used newer tools and higher-frequency data, the results aren’t identical—though they’re pretty close. The point was to test how well the classic signal holds up and to give myself a solid baseline before bringing in any extra variables.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXes7wI4kvGM4_6n3hiRCNmqhvPafqNmdniHpbKIOPn0RE1lgBwjd55GRDQVLgTi9OOApCQXro3RbWxnVkP1DpqlKXl-3YjNdyewUlLvmXmJD70uutxKmJLIo4hW0VNTzhd6096kcg?key=qhr3jCzOLCz5xmaAemL5Ng)

## Building a Broader Model: Beyond the Yield Curve

After reworking the classic Fed yield curve model, I wanted to go deeper and build something that captured more of what’s actually happening in the economy right now. The Fed’s original approach is simple, but in today’s world, recession risk is shaped by more than just bond market signals. I set out to design a model that blended the traditional—like the yield curve—with a handful of indicators that reflect the changing nature of work, debt, and public mood.

I started by creating a wide range of factors. Some were obvious, like prime-age employment (it gives a direct look at job market health) and wage growth (paychecks still drive the bulk of household spending). Others, like debt service ratios, are about how much financial stress households are feeling as borrowing costs rise. What really sets this model apart, though, is the inclusion of Google search data for “recession.” It’s a way to quantify public anxiety in real time—something you can’t get from economic statistics alone.

The process wasn’t just about piling on variables. I ran each factor through rigorous testing: dropping, splitting, and combining them to see which actually contributed useful, independent signals. Real wage growth was split out into wage growth and inflation separately, to get a clearer sense of what’s driving any change. Factors that didn’t add much predictive power or overlapped too much with others were cut.

A major limitation popped up here: adding Google Trends meant I could only use data from 2004 onward, since that’s when reliable search data starts. On one hand, this makes the model highly relevant to our digital, always-connected era, picking up on panics and sentiment swings that are more common now. On the other hand, it cuts off a big chunk of history, including some of the most important past recessions. There’s an argument that pre-2004 downturns—before smartphones, social media, and viral economic news—just aren’t as useful for understanding today’s risks. But losing that historical range also means the model hasn’t been “stress-tested” on the full variety of U.S. economic crises.

In the end, the model balances old and new: yield curve for the traditional signal, credit spreads for market stress, prime-age employment and wage growth for the real economy, debt service ratios for financial pressure, and Google Trends for public sentiment. The result is a tool that aims to be sharper and more responsive to the reality of 2020s America—even if it means giving up some of the lessons from the more distant past.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeQj4iJ0_0siUY1yVuCsmQ3X93SNXmC1qrRdA3OipCmg7feE1KblTypcU839oRX8hyoFkjtqOzUjFPyDX8q5HWx20Jp-sA03BO6XcpHQ44ndq1wTiKmRNWkXhB4opWCCC_MYZVU?key=qhr3jCzOLCz5xmaAemL5Ng)

## Testing Delayed Effects

The non-lagged model was a solid starting point, but real life isn’t that simple—sometimes economic signals take a while to actually hit the economy. Things like the yield curve and credit spreads can flash warning signs long before we see the real fallout in jobs or spending. To figure out if timing made a difference, I went back and tested what happens if you shift the yield curve and credit spreads back a few months. For the yield curve, I looked at what happens if you lag it by 6 or 12 months. For credit spreads, I tried 3 and 6 months. These aren’t random choices; they’re pretty standard in the research.

For each test, I built a new model, swapped in the lagged variable, and checked the numbers—looking at things like Pseudo R² and AIC to see if the fit improved. What stood out is that lagging the yield curve by 12 months and credit spreads by 6 months gave the best results. Basically, the warning signs from financial markets do matter, but you’ve got to be patient—those signals often take a while to show up in the broader economy.

You can see how each model performed in the graph below. The lagged models picked up on rising recession risk earlier and gave clearer signals than the versions without lags. Not every lag helped, though, and the best fit came from a 12-month lag for the yield curve and 6 months for credit spreads.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdGYY_TiXxRUZ6WkZxa-N4hfityVrBdoFeMKYTVXtteysDmQXM_I2hjkgLLwHyd0QplCuFXa-BlKGAB-in-s3mC0p1Y8A6XUyQHoWsPxyMYPpoxwCPSH6AI892PpziZCh6jBQ7wqw?key=qhr3jCzOLCz5xmaAemL5Ng)

## Looking Inside the Model

To really get a sense of what’s driving the numbers in my model, I put every key factor on one chart: wage growth, prime-age employment, debt service, yield curve, credit spread, and Google Trends for “recession.” I didn’t do this to compare the absolute values—each one has its own scale—but to watch how they actually move over time.

What I’m interested in is the story these lines tell. You can see which factors are steady most of the time and which ones swing up or down when something big happens. You start to notice if certain indicators get volatile when the economy’s shaky or if they move early before a recession or recovery. Sometimes, one factor starts to drift before the rest; other times, everything shifts at once.

The point isn’t just to spot reactions to bad news, but to see what these trends look like during good periods too. It gives a full picture: how each piece behaves when things are calm, and how the patterns change during stress. By seeing it all at once, it’s a lot easier to pick up on signals you’d miss if you only looked at one number at a time.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdB3CK5MDTs8etozkPYWLQ2-0U3f3EWvosevCV-JdKVBfoFvI32xlmJ2szAs4FRHgi8ymTSYK10uTavi86-vkooowNB1_aUUwHcFmf2jRvMzuISgDt_1vWRkPGOk-fgGJ4PuG8iyA?key=qhr3jCzOLCz5xmaAemL5Ng)

Prime-age employment has a pattern: it rose in the early 2000s, plateaued ahead of the 2008 recession, dropped sharply during the crisis, then recovered and climbed steadily. It took a sudden hit during COVID but bounced back quickly. Now it’s flattening out again, which has often signaled that the labor market is losing momentum and trouble could be brewing.

Wage growth has always bounced around, but lately, it’s been trending lower from the higher post-COVID peaks. The story isn’t one of major volatility, just a steady loss of momentum. If this keeps up, it means workers are getting less of a boost, and that can ripple through the whole economy.

The debt service ratio—the share of income going to debt payments—was high before the 2008 crash, then steadily declined for years. It fell hard during COVID thanks to stimulus and refinancing, but now it’s creeping up again. Paired with slower wage growth, it’s clear that household finances are feeling more strain.

The yield curve inverted before the 2008 recession, then returned to normal, only to take a major dive in 2022. It’s stayed deep in negative territory since then—a classic sign of economic stress that still hasn’t resolved.

Credit spreads loosely mirror the yield curve. They saw a dramatic spike in late 2008 and 2009 (reflecting real financial stress), then again during COVID. Since then, they’ve been on a slow, subtle decline, approaching the path of the yield curve. This means markets aren’t panicking, but the lack of recovery isn’t especially comforting.

Google Trends for “recession” show a clear pattern: spikes before and during downturns—like in 2008 and again with COVID—followed by quieter periods. After a huge spike in 2022–2023, search interest has bounced around. But right now, even beyond the data, the near-constant news about tariffs, Fed policy, and new economic policies is fueling uncertainty. Americans are hearing more questions and concerns about where the economy is headed, especially as debate heats up over the impact of recent policy changes.

## What’s Next?

No model settles this. Recession calls are probabilities, not predictions, and anyone telling you they know is selling something. What the model does is show you which signals are firing and which ones aren't.

Right now they don't agree. The financial market side is flashing. The yield curve has been inverted since 2022 and still hasn't resolved, and with the 12-month lag that fit best, that puts us deep inside the window where the signal has mattered historically. The real economy side is softer. Prime-age employment has flattened, but it hasn't dropped. Wage growth is slowing, but it hasn't reversed. Credit spreads are still narrow, which means markets aren't pricing in trouble at all.

The place I'd watch is households. Debt service climbing while wages slow is the combination that turns a soft patch into a real one.

So: risk is elevated, mostly on the strength of market signals that haven't shown up in jobs or spending yet.

Here's what would change my read. If prime-age employment starts falling instead of flattening, this moves from elevated to likely. If credit spreads widen, markets have caught up to the curve. Going the other way, if the curve normalizes while employment holds, the 2022 inversion was probably a false alarm. That happens, and my model can't rule it out, since Google Trends caps my data at 2004 and I never got to test it against the older recessions.

Want to see the numbers and my work behind the scenes? Check out my Github:

[https://github.com/Zetty0827/Projects/blob/main/RecessionModelling.ipynb ](https://github.com/Zetty0827/Projects/blob/main/RecessionModelling.ipynb)

Want to see the Fed's model? Click the link below:

[https://www.newyorkfed.org/research/capital\_markets/ycfaq#/ ](https://www.newyorkfed.org/research/capital_markets/ycfaq#/)


Rebuilding the Fed's Recession Probability Model

When I first started investing, the go-to answer from everyone was the S\&P 500. It wasn’t even a question—if you weren’t trading every day, you were just supposed to keep putting money in and wait. That’s what everyone said works.

But over the last few years, things changed. Tech keeps taking over, the economy feels a lot more unpredictable, and Bitcoin exploded in a way nobody really saw coming. Suddenly, the old advice doesn’t seem as solid. The S\&P’s had some rough patches, and Bitcoin, for all its craziness, has made a lot of people rethink what’s possible.

So I wanted to actually run the numbers and see for myself: If you’re not a trader, is the S\&P 500 still the smart move, or is it time to look at Bitcoin? And since the Nasdaq is full of tech names and the “next big thing,” I threw that in too.

This whole project is about cutting through the usual talk and actually seeing how Bitcoin, the S\&P, and the Nasdaq stack up—risk, returns, everything—since after the 2008 crash. I wanted real answers, not just opinions.

## What I Did

1.  Pulled end-of-quarter prices for BTC, SPY, and QQQ (Yahoo Finance). Stocks from 2009, Bitcoin from 2020.

2.  Turned prices into quarterly returns for each.

3.  Tracked how $100 would’ve grown in each, plus volatility, biggest drawdown, Sharpe and Sortino. Wanted to see returns and what you’d have to put up with.

4.  Ran t-tests, ANOVA, Mann-Whitney, and checked correlations—basically, tested if any of these are really different or just look that way.

5.  Forecasted future returns (ARIMA) and future risk (GARCH) for each asset.

6.  Ran 2,000 Monte Carlo simulations for all three—just to see what 10 years could actually look like.

7.  Checked rolling beta and rolling correlation (how Bitcoin’s relationship with stocks shifts), and ran OLS regressions to see if stuff like rates or inflation actually explains anything.

8.  Looked at worst-case scenarios with VaR and CVaR.

9.  Did an event study on the FTX collapse—just zoomed in to see how each asset reacted.

## What I learned

### Growth, Risk, and the Experience 

Compound Annual Growth Rate (CAGR): 

*   QQQ (Nasdaq): 18.3% per year 

*   SPY (S\&P 500): 13.6% per year 

*   BTC-USD (Bitcoin): 65.8% per year (since 2020)

CAGR tells you the average annual return, including compounding. Bitcoin’s growth rate is way beyond either index, reflecting how explosive it has been.

Maximum Drawdown (worst loss from peak to trough): 

*   QQQ: -32.6%

*   SPY: -23.9%

*   BTC-USD: -71.9%

Drawdown captures the pain of big crashes. Bitcoin’s drawdowns have been over double the worst drops in QQQ or SPY.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXfarCCJZr-weYKOHeVL7dfmQ8g1bwHZvzNtXROHGSTeil5f7ehLwiKLrgay3iV2etSdPrWv1oRCluf-CLpdTF5mlAF0mIuw0SjGwrLM5x0xHSCB_XS_dykouJuY2D_YU3haL5rE?key=PJsXnYpGWPzp0jbCzYmaZw)

Sharpe Ratio (reward per unit of total risk):

*   QQQ: 0.52

*   SPY: 0.48

*   BTC-USD: 0.46

Sortino Ratio (reward per unit of downside risk):

*   QQQ: 0.52

*   SPY: 0.36

*   BTC-USD: 0.88

Despite all its swings, Bitcoin’s return for the risk taken is similar to the indexes, and by the Sortino Ratio (focusing on negative volatility), it actually comes out ahead.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdVycwSgQzrgacEF5NWdyt17zqhoMCqFZmPm2U-M5jjJGPCRU4iwjoEmcgh9jwlJKKnKfjM68R2-KNtDoTR-kWmvxcu-F_WUUvWsPWXW_aqwb7txy8kg7-gLO8DKd-7IhneTimQ?key=PJsXnYpGWPzp0jbCzYmaZw)

If you invested $100 in Bitcoin and just held on, you’d have the biggest gains—but you would have had to sit through extreme ups and downs. QQQ beat SPY on both growth and risk-adjusted returns, showing the tech tilt has paid off for investors.

### Are These Differences Real? 

Investing isn’t just about eye-popping averages—you want to know if the results are reliable, or if you just got lucky. That’s where statistical tests come in.

T-test: Tests if Bitcoin’s average return is really different from the indexes, or if it’s just a statistical fluke. A p-value below 0.05 means you can be pretty sure it’s a real difference (by conventional research standards). Here, BTC vs QQQ is 0.12, BTC vs SPY is 0.10. Neither clears the 0.05 bar, so I can't call the difference significant. They're close enough to be worth noting, and with only 22 quarters of Bitcoin data, a real difference could easily fail to show up.

ANOVA: Looks at all three assets together to see if at least one has an average return that stands out. With a p-value of 0.085, it’s just shy of being “statistically significant,” but it does suggest something unusual is going on—especially with Bitcoin’s numbers.

Mann-Whitney U Test: Rather than only comparing averages, this test looks at the entire spread of returns. It asks: Do these assets give you the same kind of ride, or are they fundamentally different in how gains and losses show up?

The results came back at p = 0.50 against QQQ and p = 0.45 against SPY, so by this test I can't tell the distributions apart. That surprised me given how different the volatility numbers look, and I think it's mostly a sample size problem. Twenty-two quarters of Bitcoin data isn't much. The volatility and drawdown numbers below make the difference clearer than the significance tests can.

### Do These Assets Move Together?

Correlation measures whether assets move in sync or not. A value of 1 means they move together perfectly; 0 means they move totally independently. ​​Beta measures how much an asset moves compared to a benchmark—here, it’s “rolling,” meaning it’s recalculated over a moving window so you can see how relationships shift over time.

*   QQQ and SPY: 0.92 (very high—almost always move in the same direction)

*   BTC-USD and QQQ: 0.47

*   BTC-USD and SPY: 0.52

![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXc_GUjd-uK4FBr_2cYYRqNIhbOgL-2y94VAQTFa_vajXUM7HRgNm8YNEeP4HomnOJ3_Tht42gklkQdB40jhbhHYNI5ngxYPmRWXHmty3ccUmneaFbwgyaYtKl8fDZ7p-wZpv4og6A?key=PJsXnYpGWPzp0jbCzYmaZw)This graph tracks how closely Bitcoin’s returns move with the SPY and the Nasdaq (QQQ), quarter by quarter. The lines show the rolling correlation and the rolling beta (how sensitive Bitcoin is to moves in each index). The correlation lines vary, but move around 0.5. This means Bitcoin sometimes moves with the stock market. When beta is above 1, Bitcoin usually moves more than the S\&P or Nasdaq. So, if beta is 1.5, and the S\&P 500 rises 2%, you’d expect Bitcoin to rise about 3% (or fall 3% if the S\&P drops 2%), on average.

Bitcoin only loosely tracks the stock indexes. Sometimes it rises when stocks fall, and vice versa. That means adding Bitcoin to a traditional portfolio can offer real diversification, sometimes softening the blow during stock selloffs—but it can also make things bumpier overall.

### How Bad Can It Get?

Volatility tells you how much an asset’s value jumps around each quarter. Value at Risk (VaR) and Conditional VaR (CVaR) are ways to size up the worst-case scenarios.

Quarterly Volatility (Std Dev):

*   BTC-USD: 54%

*   QQQ: 9%

*   SPY: 7%

5% Value at Risk (VaR):

*   BTC-USD: -39.2% (5% chance of losing at least this much in a quarter)

*   QQQ: -10.0%

*   SPY: -13.2%

CVaR (Average of Worst 5% Quarters):

*   BTC-USD: -48.5%

*   QQQ: -15.1%

*   SPY: -15.7%

One thing that looks off but isn't: QQQ's 5% VaR is shallower than SPY's, even though QQQ is more volatile and had a deeper max drawdown. That comes from the window. Starting in mid-2009 cuts out the 2008 crash, and since then QQQ's bad quarters have been milder than the S\&P's while its good quarters have been much better. More volatile, but the swings lean upward.

![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXc-whDjkc0sCfhJUZ8X3VgJoHIQsAORBcOdwus985AOarJrepv0OU--cHEvNas4czE0lbfO-rUNsfWc1_fxe6JcGxKzWPZrRKL4oWkm-hRng6Oj2HWS3yMaaak5NEs3uquHyNVY5Q?key=PJsXnYpGWPzp0jbCzYmaZw)

This graph shows the volatility over time for each of the assets. Both the S\&P and NASDAQ have low and fairly stable volatility. Bitcoin on the other starts off much higher, but is on a path towards stabilization.

### What Drives These Returns?

I ran OLS regressions to see if common factors like the market, interest rates, or inflation can explain how these assets move.

*   SPY: R² = 1.00 (this one is circular, since SPY is the market factor, so I'm leaving it as a reference point rather than a result)

*   QQQ: R² = 0.82 (market and tech explain most of the moves)

*   BTC-USD: R² = 0.32 (not linked to broad economic factors)

R²: A high R² (like QQQ) means the asset's returns are predictable from big-picture trends. A low R² (like BTC) means a lot is happening that standard models can't explain.

### What Might Happen Next?

![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdBTGzl-T_jYdNHJk0A6_dyjajGz4NkQsjWs1g7BJR1cZfcJs-5M5Vgn2gzq2bE3yIwdVRM267gS4BFAbczT7bdsDYStEP8hdc4j4hTIUVz9qMpaO4u_oNyigMJS335r_SpBjG9pQ?key=PJsXnYpGWPzp0jbCzYmaZw)ARIMA models future returns based on past trends and cycles. For SPY and QQQ, the forecast lines are smooth and steady—they’re expected to keep producing similar returns to the past. For Bitcoin, the ARIMA forecast shows high expected returns despite its volatility.![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXfjQDzouQfNT2ZaWCHt6ZS73hAobwn3f3fCxCi5YtKnpP0lW3HBoTGjA_FSOhluncmW17dgGQQN_FC4Hv8-Jhi3fOkKUrUJRLQTJ--MmANu3GxTI0EfXGblZ3GAFJvIWFp1R_xK?key=PJsXnYpGWPzp0jbCzYmaZw)GARCH models forecast how risky (volatile) each asset will be. Stocks are projected to stay low-risk, with volatility barely moving. Bitcoin’s volatility, despite being much higher than the others, is on a downtrend.

### Monte Carlo Simulation: Many Possible Futures

![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdJypQPbkhKvx2SSPnh5bgZYpBgxw_h_loV31y3pXvA054WwqO6nzmADmVIejWNLp39ohyY-ZsLS-ovBIPFv0m3wDJ2QWRc29Ijs5HNbnZs39wy2lnTH5Tp-vIss891KYZxT1E4xw?key=PJsXnYpGWPzp0jbCzYmaZw)

A Monte Carlo simulation doesn’t just give you an average—it shows you the range of what could actually happen if you invested today and held for 10 years, using the real ups and downs from the past.

*   SPY: Most scenarios are steady and positive—you rarely lose big.

*   QQQ: Slightly more spread; higher gains are possible, with some extra risk.

*   BTC-USD: The widest range by far—high concentration of simulations show profit, some even 10x or 100x, while others have you crashing to zero or close to it. 

This isn’t about predicting a single future, but showing the real odds: With Bitcoin, the chance for outlier success is huge, but so is the risk of disappointment. Stocks offer much more predictability, but not the same shot at outsized gains.

### Stress Test: FTX Collapse and Real-World Shocks

![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXcdkQ1RCsNG0e0b8xWxmea1dM21bbST2WrFhjjQN1Neu4UVcyBxXh4EN5pLg3sZV8bcJPp925Jgm9JljCOgfN2AXXONCwqiISVhxrpcxlv2vTetgB75yEbMPA0Zy5lHFoIQ2BT-gg?key=PJsXnYpGWPzp0jbCzYmaZw)The FTX collapse in late 2022 was a huge moment for crypto. When I zoomed in, Bitcoin, SPY, and QQQ all dropped—but none of the drops were extreme for their own histories. For Bitcoin, even a collapse like FTX is “just another bad quarter.” The t-stat was -0.31 and the p-value was 0.77.

A p-value below 0.05 is the benchmark for calling a result “significant”—it would suggest the drop was unusual compared to the past. A p-value of 0.77 is much higher, which means the drop during the FTX event was well within the normal range of Bitcoin’s typical volatility. For both stocks and Bitcoin, the event wasn’t an outlier when measured against their own wild histories.

## Limitations

*   We don't have much Bitcoin data—just since 2020, which works out to 22 quarterly observations. That means a few big quarters can swing the whole comparison, and it's thin for the GARCH and ARIMA forecasts in particular. Those models usually want a lot more history than that.

*   The stock indexes start in 2009, so the periods don’t really line up. We’re not seeing how these assets behave through the same cycles.

*   By using quarterly returns, we’re smoothing over the rough stuff that actually happens month to month, especially with crypto.

*   This is just looking at each asset alone. No mixing, no rebalancing, no “real world” portfolio building.

*   None of the numbers reflect trading fees, slippage, or taxes. Especially for Bitcoin, those can eat into returns.

*   By setting the risk-free rate to zero, we slightly inflate the Sharpe and Sortino ratios. Using the real US Treasury rate would lower them a bit, especially if rates are high. This doesn’t usually change the overall ranking, but it does mean the ratios aren’t exactly what you’d see in a professional fund analysis.

*   All the forecasts (ARIMA, GARCH, Monte Carlo) are just projections based on what’s happened lately. If markets change in a big way, these models won’t catch it.

*   Indexes like QQQ and SPY drop underperformers, so long-term results are boosted by survivorship bias.

*   The math doesn’t show what it feels like to live through big drops or wild swings. It assumes nobody sells during the tough times—which isn’t how most people actually behave.

## Takeaways

1.  The S\&P 500 isn’t the gold standard it’s made out to be. The data shows that the Nasdaq (QQQ) has left the S\&P behind, both in growth and risk-adjusted returns. Tech-driven stocks have been where the real gains happened.

2.  Bitcoin's returns are real, but the risk-adjusted case is weaker than the headline. The raw growth is unmatched at 65.8% a year since 2020. But the Sharpe ratio comes in at 0.46, below both QQQ and SPY, so you're not actually getting paid more for the risk you take. Sortino is friendlier at 0.88, so if you only care about downside swings it looks better. The volatility trend is heading down, which is the most interesting thing in the data. Five years isn't enough to say whether that's a trend or a phase.

3.  The diversification argument is the stronger one. Bitcoin's correlation to stocks sits around 0.5, and market returns, rates, and inflation together explain only about a third of how it moves. Most of what drives Bitcoin isn't in the standard models. That independence is what makes it useful in a portfolio, more than the returns. It also means the usual tools can't tell you much about where it goes next.

4.  Personally, I’m moving away from buy-and-hold S\&P 500 investing and focusing more on the Nasdaq for long-term growth. I will keep investing in crypto, but I’ll be watching closely—potentially using active strategies to take advantage of its still-immature market dynamics.

Want to see what I did behind the scenes? Check out my Github

<https://github.com/Zetty0827/Projects/blob/main/Crypto_VS_NASDAQ.ipynb>


Comparing Bitcoin to the S&P and the NASDAQ

Technology isn’t a tool anymore — it’s infrastructure. It’s how we live, work, move, pay, speak, and exist. You need it to function in modern society. And the more we depend on it, the more dangerous it becomes when that access is conditional.

Can you get a job without internet? Buy groceries without a debit card? Apply for housing without an email address? In some cities, you can’t even get the cheapest bus fare without a smartphone. Across the world, governments and businesses are moving away from cash. In Sweden, cash is used in less than 9% of daily transactions. In India, mobile payment apps dominate everyday commerce. In cities like New York and San Francisco, card-only establishments are becoming the norm. The message is clear: if you’re not digitally connected, you’re increasingly invisible.

And yet, the systems that mediate that access are privately owned, and they answer mostly to the people who own them.

Let’s talk about EULAs — End-User License Agreements. They’re the fine print you blindly accept to use your phone, your bank app, your cloud storage — basically, your life. Most people don’t read them. That’s not an accident. These documents are written to shield corporations from responsibility, not to protect users. The terms can change at any time, and if you disagree, you lose access.

Apple’s recent capitulation to the UK government proves the stakes. Under pressure from the UK’s Investigatory Powers Act, Apple disabled its Advanced Data Protection feature — a setting that gave users end-to-end encryption of their iCloud backups. It only applied to UK users, but the precedent is chilling: when power knocks, privacy folds. And you’re never consulted.

What happens when the thing you rely on to store your memories, access your accounts, and communicate with the world can be turned off, just like that?

This isn’t limited to cloud storage or photos. It’s happening in finance, too.

Globally, 1.4 billion people still live without a bank account. In the U.S., nearly 6 million households are unbanked, and millions more are “underbanked,” relying on expensive workarounds like payday loans or check-cashing services. And it’s not because they’re irresponsible. It’s because they’ve been priced out, filtered out, or ignored.

Regulations like KYC (Know Your Customer) and AML (Anti-Money Laundering) are supposed to prevent fraud and criminal activity. But in reality, they often block people without stable documentation, like unhoused individuals, immigrants, or the poor, from opening accounts. Onboarding a customer has real compliance costs, and a low-balance account rarely covers them, so banks have every incentive to exclude.

Fintech was supposed to change that. Instead, it put the same barriers behind prettier screens.

Now your access to money depends on whether your app is compatible, your phone is charged, your selfie matches your ID, and your behavior doesn’t trigger a security flag. You’re not onboarded — you’re screened. And once you’re in, you’re monitored.

Every transaction, every search, every location ping becomes a data point. Platforms say it’s to “improve your experience,” but in practice, it’s to predict you, sell you, and shape you. The moment you enter these systems, you become the product. Your habits, your preferences, your trustworthiness — all calculated and packaged for monetization.

And if your behavior becomes inconvenient to the platform — politically, financially, or socially — your access can be throttled, suspended, or erased. You don’t have to break the law. You just have to become unprofitable.

This is how people get digitally exiled — not for crimes, but for misalignment. This is how control works in the age of infrastructure.

There's an old answer to this problem, and it's worth remembering. When railroads, telephone lines, and electric grids became things you couldn't live without, we didn't leave access up to whether you were profitable to serve. We treated them as common carriers and utilities and attached obligations to them. Serve everyone at posted rates. Don't discriminate. Don't cut people off without some kind of process. Universal service funds paid to wire rural areas that no company would have wired on its own.

Digital payments and identity haven't gotten that treatment, mostly because they didn't look like infrastructure when the rules were written. They do now.

Milton Friedman, who was no fan of regulation, argued that when you're dealing with monopoly you're choosing between private monopoly, public monopoly, and public regulation, and that all three are bad. You pick the least bad one. That's the honest version of this problem. Leave it alone and exclusion gets priced in. Regulate it and you get new costs and new ways for the rules to be captured.

There are real options on the table: a basic account that doesn't require perfect paperwork, KYC requirements that scale with the risk of the account instead of applying the same checks to everyone, interoperability rules so switching providers isn't a trap, and terms of service short enough to actually read. Each one trades something away. But they're all answers to the same question we already answered for water and electricity: once something becomes a condition of participating in society, who's responsible for making sure people can get it?



The Infrastructure of Control

It's taken Google decades to build an empire in search, advertising, and digital services, shaping much of the web. For most businesses and customers, it's not possible to circumvent Google—it's everywhere online. As Google's influence grows, regulators are asking whether it promotes competition or just reinforces its monopoly. The new claims are a result of an EU investigation after rival companies claimed that Google's dominance of search and mobile platforms was stifling legitimate competition. In March 2025 the European Commission issued preliminary findings that Google was favoring its own products in search and blocking app developers from steering users to better deals outside Google Play.

At its core, the allegation is Google biases its search results to promote its own services, such as Google Flights, Google Hotels, and Google Shopping, over those of competitors being pushed further towards the bottom of the page. This isn't the first time—Google has been accused in this manner previously. Google was fined €2.42 billion by the EU in 2017 for prioritizing its shopping service over others. That ruling affirmed a clear precedent: Google dominates virtual marketplace access, and it exercises that control to maintain its dominance (Reuters, 2025). Another major issue is Google's Play Store restrictions. The EU's concern there is that Google stops developers from telling users about cheaper offers outside the Play Store, and charges them fees even when they do. That keeps transactions inside Google Play, where Google takes its cut (Financial Times, 2025).

Google's scale is a big part of the story, but it isn't only economies of scale. It's network effects. More searches give Google more data, more data makes its results better, and better results bring more searches. A competitor starting from zero can't get into that loop.

1.  Search dominance: Google handles an estimated \~90% of European search traffic, according to the report. Competitors can't match the volume of data behind its results, and that gap is hard to close no matter how good their product is (Wall Street Journal, 2025).

2.  Advertising muscle: Google operates one of the world's largest ad networks, making it a digital advertising giant, particularly in search ads.

3.  App Store domination: Google takes up to a 30% commission on Play Store transactions, with a lower 15% rate on a developer's first $1 million in annual revenue and on subscriptions.

Google's side of this isn't empty. Integrating Flights or Shopping into search gives users faster answers, and plenty of people prefer that to clicking through to a comparison site. The hard question for regulators is separating self-preferencing that shuts out rivals from integration that people actually want. That line is genuinely difficult to draw, and it's why these cases take years.

If the EU wants the DMA to work, then it needs to do more than just fine companies. Whether or not enforcement will matter is still uncertain. Without better enforcement, Google will simply continue to look at fines as merely another cost of doing business.

A possible solution would be to require Google to unbundle its search and ad businesses in an attempt to prevent self-preferencing. This would create a clearer divide between how it ranks searches and how it profits from them. Another step would be forcing Google to allow genuine third-party app stores on Android without restrictive policies, giving developers and consumers more choices. Regulators could also require greater transparency in Google's search ranking system to prevent favoritism for its own offerings.

The bigger risk is that the rules stay vague. Ambiguous obligations give a company with Google's legal resources room to comply on paper, delay, and relitigate, while regulators play catch-up. If the DMA is going to work, the obligations have to be specific enough that compliance is something you can actually check. Otherwise the fines just become part of the budget.


Regulating Big Tech: Will the EU Finally Rein in Google?

Technology has always advanced economies and societies, but technology is now attempting to control them. In a recent interview, social psychologist Jonathan Haidt argues that platforms like Facebook, Twitter, and TikTok haven’t just transformed how people interact—they’ve shifted power away from traditional institutions and concentrated it in the hands of a few companies. Haidt’s argument goes beyond social media’s effects on mental health or polarization. He sees it as something larger, a restructuring of economic and political control. The way information spreads, the way businesses compete, and the way governments function are all being rewritten by algorithms designed to maximize engagement, not serve the public good.

Haidt points out that social media companies operate differently than traditional businesses. Instead of selling a product or service, they sell human attention. Every click, comment, and interaction fuels an industry where user behavior itself is the product. These companies don’t just profit from what people do; they shape it. Competition in this space barely exists. When a platform gets too powerful, it doesn’t face challengers—it buys them out or buries them. Facebook swallowed Instagram and WhatsApp. Google controls search, advertising, and video content through YouTube. These companies don’t just dominate industries; they define the rules of engagement. Meanwhile, institutions that once held authority—news organizations, universities, even governments—now depend on social media to function. Information is filtered through algorithms that no one outside the tech industry fully understands, let alone regulates. The economy hasn’t just adapted to social media; it revolves around it.

Haidt argues that social media, driven by algorithms, shapes people's opinions by determining what they see. Algorithms don’t reward accuracy; they reward engagement. The posts that trigger the strongest reactions—whether through outrage, controversy, or pure entertainment—are the ones that spread. This shift has weakened institutions that once held authority over knowledge. Universities, journalists, and policymakers no longer compete on credibility but on visibility. The algorithm, for example, determines whether a well-researched report or a viral conspiracy theory gets more visibility. But Haidt’s concern goes beyond misinformation. Social media doesn’t just reflect user behavior; it trains it. Platforms guide interactions by pushing certain types of content while suppressing others. They don’t just respond to human habits; they shape them, reinforcing behaviors that keep people engaged and filtering out those that don’t. The result is an environment where what gets attention isn’t what’s most valuable—it’s what holds the user’s gaze the longest.

Haidt claims tech monopolies have taken power away from governments. Traditional regulatory systems fail to keep up. Laws written for print newspapers and television networks don’t apply in a world where algorithms determine what people see. The European Union has attempted to rein in digital monopolies, but in the United States, regulation is almost nonexistent. Without intervention, Haidt argues, these companies will continue to shape the economy, society, and possibly even democracy itself.

Haidt references Four Futures: Life After Capitalism by Peter Frase, a book that outlines four possible directions for society based on economic and technological trends. The four scenarios are:

1.  Communism – Technology eliminates scarcity, leading to a world of shared prosperity.

2.  Rentism – A few corporations maintain control by owning intellectual property and digital infrastructure.

3.  Socialism – Scarcity forces society toward a collectivist economy.

4.  Exterminism – A small elite hoards resources while the rest of society struggles to survive. 

Haidt believes we are moving toward rentism.

Haidt’s concerns aren’t just speculation—they’re already playing out in ways that are easy to overlook. Some tech companies don’t just dominate markets; they build their own ecosystems. Google’s headquarters provides nearly everything employees need on-site, Meta is developing "Willow Village," a company-run neighborhood, Amazon’s warehouse workers rely on company-provided transport and logistics, and Apple’s $5 billion campus keeps employees inside a closed-loop work environment. Workers live and operate within corporate-designed spaces, cut off from the economic realities outside, while the cities around them deal with rising costs and homelessness. These companies operate as if they exist outside the societies they profit from.

We haven’t reached a point where regulation is impossible, but waiting too long will make it harder to reverse course. These companies are embedding themselves deeper into economic and social structures, making them more resistant to change. If they continue expanding without oversight, the window for meaningful intervention will shrink. Stronger antitrust enforcement, transparency in algorithmic decision-making, and stricter data protections are necessary to curb their growing influence. This isn’t just about market competition anymore—it’s about who decides what the future looks like.


Blog

Rebuilding the Fed's Recession Probability Model

Comparing Bitcoin to the S&P and the NASDAQ

The Infrastructure of Control

Regulating Big Tech: Will the EU Finally Rein in Google?

Technology, Economy, & Society