Like many in the Bay Area, I've had the good fortune of receiving RSU compensation at a large, growing public company, and as such I've had to consider the high-class "problem" of whether and how to diversify a concentrated single-stock position. People approach this in different ways: on one end of the spectrum, you have the financial advisor types who push maximum diversification and on the other, the HODL'ers who are in a mild panic every time the company stock has a bad spell.
I've heard a variety of approaches to diversification from others, but often it's pretty ad-hoc, selling chunks of stock here and there and otherwise using employer stock as a piggy-bank for larger cash needs. I wanted an approach that was more principled, and I eventually settled on a pretty simple rebalancing method. I've sometimes wondered how well that approach performed in terms of total and risk-adjusted returns, and this project attempts to answer that question.
Not investment advice. This is a historical simulation for curiosity and learning. Past performance does not predict future results, and the model omits real-world details such as wash sales, tax brackets, dividends as cash, transaction costs, and slippage.
Each year, I would receive an annual refresher grant of RSUs denominated in dollars, roughly proportional to my salary. The grant's dollar value would convert into a share count at the grant date, and then those shares would vest over the following four years. Because the share count is locked at grant, the dollars delivered at each vest float with the share price. If the stock has appreciated, the grant over-delivers, which is wonderful but concentrates you even further in employer stock.
My diversification approach was a one-way threshold rebalancing strategy. Once or twice per quarter — shortly after the trading window opened, and once more (if I remembered) just before the next blackout period — I would calculate what fraction of my total stock holdings were in employer stock. If that exceeded a threshold (e.g., 33%), then I would sell my excess employer stock down to that threshold and buy a diversified index with the proceeds. It was a one-way rebalance only; since I expected future vesting, I would only ever sell employer stock, never buy it.
It's a simple strategy, but I found it had some nice benefits. First and most importantly, my selling decisions were entirely mechanical. I did a quick calculation, and then I sold a specific amount of stock, which took much of the emotion and FOMO out of it. Second, like any rebalancing approach, the strategy naturally enforces buy low, sell high discipline. You sell more stock when your employer outperforms the market and you sell less (or hold) when it lags. In practice, annual vesting would almost always trigger a sale, and in between, the rebalancing rule would trigger a sale if my employer's stock had significant appreciation relative to the broader market.
This project backtests the one-way threshold rebalancing strategy. It is built as a small Python package for the backtest engine plus an interactive marimo notebook front-end.
The backtest engine runs a strategy through the desired time window to calculate risk and return characteristics of the strategy. The engine implements a simplified model of vesting schedules as well as taxes upon vesting and sale events. It idealizes many other details; for example, it assumes instantaneous conversion from employer stock to index with fractional stock and zero transaction costs.
The output of the backtest compares three strategies:
| Strategy | What it does |
|---|---|
| Hold everything | Never sell — maximum concentration. |
| Threshold N% | The target strategy — trim employer stock to N% on each rebalance day. |
| Sell all at vest | Convert every grant straight to the index — full diversification. |
Both runs below use the notebook defaults — VTI as the diversified index, 2019–2024 date range, the same nominal grant values and tax rates — and differ only in the employer stock. In the figures, each line is the time-weighted return under one strategy, and the shaded bands mark stretches where the index outperformed the employer.
Because grants add money over time and tax payments behave as withdrawals, raw portfolio value is
not a clean return series — a jump on a grant day is a deposit, not market performance. Risk and
return here use a time-weighted return that strips out each day's non-market inflows and
outflows (metrics.time_weighted_returns). Final dollar value is still reported directly and is a
fair head-to-head number because every strategy receives the identical grant stream.
| Metric | Hold everything | Threshold 33% | Sell all at vest |
|---|---|---|---|
| Final portfolio value | $1,705,043 | $1,282,337 | $1,114,918 |
| Liquidation value (net of tax) | $1,417,665 | $1,113,975 | $997,484 |
| Vested contributions (net of tax) | $705,891 | $705,891 | $705,891 |
| Taxes paid | $554,628 | $561,791 | $554,628 |
| Annualized return (TWR) | 34.76% | 21.71% | 14.13% |
| Annualized volatility | 30.25% | 22.40% | 20.24% |
| Max drawdown | -31.43% | -33.46% | -35.00% |
| Sharpe ratio | 1.07 | 0.90 | 0.66 |
| End employer % | 100.0% | 36.7% | 0.0% |
Over this window AAPL handily beat the market, so holding everything won outright on final dollars, but it rode the most volatility and the deepest drawdown to get there. Even for a high-flying stock like AAPL, there were periods when the stock underperformed the index. The 33% threshold strategy gave up much of the single-stock upside in exchange for diversification and downside protection and came out ahead of full diversification on both raw and risk-adjusted return.
| Metric | Hold everything | Threshold 33% | Sell all at vest |
|---|---|---|---|
| Final portfolio value | $173,390 | $466,358 | $652,068 |
| Liquidation value (net of tax) | $173,390 | $418,165 | $576,499 |
| Vested contributions (net of tax) | $389,578 | $389,578 | $389,578 |
| Taxes paid | $306,097 | $306,159 | $306,097 |
| Annualized return (TWR) | -13.00% | 6.63% | 14.13% |
| Annualized volatility | 40.84% | 23.84% | 20.24% |
| Max drawdown | -69.57% | -35.03% | -35.00% |
| Sharpe ratio | -0.18 | 0.31 | 0.66 |
| End employer % | 100.0% | 16.1% | 0.0% |
Intel is the mirror image: over the same window it lagged the market badly, so concentration was punished instead of rewarded. Full diversification wins here, and the threshold strategy keeps most assets protected by diversification while holding a slice of employer exposure.
The charts and tables above are regenerated from the notebook defaults by assets/generate_assets.py.
Requires uv.
uv sync --extra devuv run marimo edit notebooks/rsu_backtest.pySet the employer/index tickers, date window, grant size and growth, and threshold to run a strategy. The backtest re-runs reactively as you edit settings to produce charts (employer concentration over time, growth of the three strategies), a return/risk comparison table, and the threshold strategy's trade log. The various settings exposed in the notebook are documented below.
from rsu_rebalancing import (
BacktestConfig,
GrantConfig,
StrategyConfig,
comparison_table,
run_backtest,
)
strategy = StrategyConfig(employer_ticker="AAPL", index_ticker="VTI", threshold=0.33)
grants = GrantConfig(grant_dollars=100_000)
backtest = BacktestConfig(start="2019-01-01", end="2024-12-31")
results = run_backtest(strategy, grants, backtest)
print(
comparison_table(
results,
risk_free_rate=backtest.risk_free_rate,
)
)The above Python code mirrors the notebook's AAPL defaults and should reproduce the summary table shown above.
These are the knobs the notebook exposes, in the same two groups it uses: the everyday ones up top, the fussy details under "Extra settings". Each maps to a field on one of the config dataclasses for the Python API. For additional details and a few fields that are not exposed to the notebook, see the dataclass docstrings in src/rsu_rebalancing/config.py.
- Employer / index ticker (
StrategyConfig.employer_ticker,index_ticker): The concentrated stock and the diversified fund its proceeds buy. - Start / end date (
BacktestConfig.start,end): The backtest window. - First-year grant $ (
GrantConfig.grant_dollars): The gross (pre-tax) award value in the window's first year, priced into a share count at the award date. - Grant growth %/yr (
GrantConfig.grant_growth_rate): Annual nominal growth of the award value, modeling wage inflation; compounds off the window's first year. - Rebalance threshold % (
StrategyConfig.threshold): Target maximum employer fraction; rebalances trim employer stock down to it.
- Backfill grants before window (
GrantConfig.backfill): Whether year one of the backtest opens with grants already vesting (a mature employee) or starts from zero (a new hire ramping up). - Vesting years (
GrantConfig.vesting_years): Equal annual tranches each award vests over. - Rebalances per quarter (
StrategyConfig.rebalances_per_quarter): How many evenly spaced rebalance dates to place in each quarter. - Hysteresis band % (
StrategyConfig.rebalance_band): A one-way band; a rebalance fires only when the employer fraction exceedsthreshold + band. - Vest withholding % (
TaxConfig.ordinary_income_rate): Ordinary income tax rate for each vesting event via sell-to-cover, grossing the kept share count down. - Short- / long-term cap-gains tax % (
TaxConfig.short_term_rate,long_term_rate): Tax rates for gains on employer-stock sales by holding period. - Risk-free % (for Sharpe) (
BacktestConfig.risk_free_rate): Used only for calculation of the Sharpe ratio.
Share prices come from yfinance with an adjusted close, so splits and dividends are handled. Fetches are memoized in-memory for the session, so that moving a slider in the notebook doesn't re-hit the network.
uv run pytest -q # tests (no network needed)
uv run ruff check . && uv run ruff format --check . # lint + format
uv run pyright # type checkSee CLAUDE.md for the project layout and conventions.
MIT