Stop guessing.
Start using data.

Daily trading signals powered by machine learning. Updated every market day. Paper trading track record you can verify.

See the signals

Three systems. Pick what fits your accounts.

Each runs independently. Subscribe to one, two, or all three.

TSP Fund Switching

For federal employees

A daily signal that tells you whether to be in C/S funds or G fund. Two ML layers: one detects when to move to safety, the other picks the best stock fund when conditions are clear.

Respects the 2-IFT-per-month limit
Backtested 2014–2025
6 ML models retrained daily
28.6% CAGR
Strategy B backtest vs 10.6% static Lifecycle (2014–2025)
View TSP Signal

Stock Selection

For taxable brokerage accounts

Automated S&P 500 stock picking. The system ranks 500 stocks daily using momentum, quality, and value metrics, then builds and rebalances a concentrated portfolio.

6 strategy variants to match your risk tolerance
Automated execution via Alpaca
Crash protection + risk overlays built in
6 variants
Paper trading live since Jan 2026 — $100K each
View Paper Trading

IRA Optimization

For tax-advantaged retirement

Same stock ranking engine, tuned for tax-advantaged accounts. No wash sale constraints, no tax-lot harvesting overhead. More aggressive rebalancing because taxes don't eat your gains.

6 IRA-specific variants
Optimized for Roth / Traditional IRA
Walk-forward validated: trained on the past, tested on the future, repeated 12 times
~6.5%/yr
Tax drag eliminated vs taxable account — compounds every year
View IRA Portfolios

How it works

1

Data comes in every market day

At 4 PM ET, the system downloads stock prices, economic indicators, volatility data, and fund performance. Everything updates automatically.

2

ML models assess risk and rank stocks

Machine learning models classify market conditions (bull, bear, neutral), estimate crash probability, and rank S&P 500 stocks by expected performance. For TSP, separate models predict which fund will outperform.

3

You get a clear signal

For TSP: “Stay in C Fund” or “Move to G Fund.” For stock portfolios: a specific list of stocks to hold with position sizes. The model tells you what it recommends and why — you decide what to do with it.

4

Track record is public

Every signal is logged. Paper trading portfolios run with virtual money so you can verify performance before committing real capital. Backtests use walk-forward validation — no cherry-picking.

Built on verifiable results

Every claim on this site is backed by data you can inspect. Backtests include taxes, transaction costs, and realistic execution assumptions.

15
Portfolios paper trading live (6 NET + 3 NET-in-IRA + 6 IRA)
12
Walk-forward validation windows
10K
Monte Carlo simulations per variant
Daily
Model retraining frequency

Why this exists

I'm a federal employee with 20 years of post-PhD experience in data science and machine learning. During the 2025 furlough I started asking what I'd actually do with my TSP if markets dropped the way they did during COVID. I had no plan, and nothing I found online was useful — just Reddit tips and analysts making gut-feel calls.

So I built my own. First for TSP, then for my taxable and IRA accounts. The system runs every market day, retrains its models daily, and logs every decision. This site is how I share it.

Frequently asked questions

Is this investment advice?
No. This site provides quantitative signals and data only. Nothing here is a recommendation to buy, sell, or hold any security. You make your own decisions with your own accounts. The signals are outputs of statistical models — they can be wrong.
Do you have access to my TSP / brokerage account?
No. The TSP signals are information only — you log into tsp.gov and make transfers yourself. For stock portfolios, automated execution is available through Alpaca (a separate brokerage), but you control your own Alpaca account and can disconnect at any time.
How accurate is the TSP switching model?
The TSP system has two strategy variants. Strategy B (the high-conviction switcher) only triggers transfers on strong signals and achieved 28.6% annualized returns in backtesting (2014–2025) vs 10.6% for a static Lifecycle allocation. Strategy C (the conservative aggregate) waits for sustained signals over 15 trading days before recommending a move. But backtests are not guarantees. The model will make wrong calls — the value is in getting more right than wrong over time. The dashboard shows full backtest details, Monte Carlo simulations, and model accuracy metrics so you can judge for yourself.
What happens when the model is wrong?
It will be wrong sometimes — every model is. The crash prediction system has a known false positive rate, and the fund selector won't always pick the best fund. The system is designed so that wrong calls cost less than right calls earn. Backtest results include all losing periods and maximum drawdowns so you can see worst-case scenarios before committing.
What about the TSP 2-transfer limit?
TSP allows 2 interfund transfers per month (additional transfers can only move money into G fund). The model is designed around this constraint. Strategy B only triggers switches on high-conviction signals, and Strategy C aggregates signals over 15 trading days before recommending a move. Neither strategy will burn your IFTs on weak signals.
What returns should I realistically expect?
Backtested CAGRs vary by strategy — the TSP Strategy B backtest was 28.6% (vs 10.6% static Lifecycle), and the stock variants range from low-double-digit to mid-twenties depending on risk tolerance. Live results will differ. Paper trading is running now so you can see actual signal-to-execution performance before committing real capital. The honest answer: assume live returns will be lower than backtests, judge the system on consistency and drawdown control rather than headline CAGR.
How is this different from a robo-advisor like Betterment or Wealthfront?
Robo-advisors run static, low-turnover allocations — they pick a target stock/bond mix and rebalance to it. They don’t time markets, don’t pick individual stocks, and don’t react to crash signals. This system does all three: ML models classify market conditions daily, rank stocks individually, and shift exposure when crash probability rises. Different philosophy — active and signal-driven vs. passive and rules-based.
How do I know the backtests aren’t curve-fitted?
Three protections. First, walk-forward validation: models train on past data and are tested on the next out-of-sample window, repeated across 12 different time periods. Second, Monte Carlo simulation: 10,000 randomized scenarios per variant to stress-test parameter sensitivity. Third, paper trading: every signal is being executed live in real time and logged publicly. Curve-fitting shows up fast in those three layers.
Is this free?
The dashboards are currently open while the system builds its live track record. Pricing will be introduced once paper trading has run for a sufficient period. Early users will get favorable terms.