Collegica Collegica Collegica
  • Start here
  • Campus
  • Apps
  • Events
  • Talks
  • About
  • Sign in
  • Newsletter

Buy the Drop, Sell the Bounce

How a TFSA returned 75% on the money put in between 2017 and 2021: 244 trades, one repeated pattern, and what the record says about skill, luck and risk

A first-person case study of a Canadian TFSA that traded a short list of large US technology stocks from 2017 to 2021: buy after a sharp drop, add if it keeps falling, sell the rebound, never sell at a loss. The trades, the rules read back from them, the return against the S&P 500 and Nasdaq-100, the cost of currency conversion, a rule-following backtest, and where the approach was exposed.
Author

Behzad Samadi

Published

November 21, 2026

A paper-craft valley of stacked navy, teal and cream paper layers shaped like a price drop and rebound, with navy candlesticks falling down the left slope and amber ones rising up the right, and a small amber ball bouncing at the bottom. Title: Buy the Drop, Sell the Bounce. Four years of trades, read back honestly.

Between January 2017 and March 2021 I paid money into my Questrade TFSA and took out 75% more than I had put in, with the account left essentially empty. That gain came from 244 trades that followed one repeated pattern: buy a familiar stock after a sharp drop, add if it keeps falling, sell the rebound a few weeks later, and never sell at a loss.

I have not published this before. I am publishing it now because the record is complete, because it is easy to tell a flattering story about it, and because the honest version is more useful. The strategy as I remembered it was “only trade well-known companies, and buy when a sudden dip happens.” The trades bear that out about four times in five. What the memory leaves out mattered as much: the quick exit, the averaging down, and one speculative side bet that produced more than a third of the gains.

This is what happened to one account in one period. It is not a method to copy, and nothing here is advice.

NoteIn short
  1. The account made money, and it made it mostly in one year. 2020, the March crash and its recovery, produced about 60% of all realised gains.
  2. The rules were never written down. Read back from the trades, they come to eight: a short list of large US growth stocks, buy after a drop, add on further falls, sell about 6% above average cost, no stop-loss, put the proceeds straight back to work, cash out after a strong run.
  3. Two stocks carried the total. Tesla and one small speculative company, Barnes & Noble Education, produced about two-thirds of the gains.
  4. Measured like a fund, it grew 27.7% a year in Canadian dollars, against 14.1% for the S&P 500 and 26.5% for the Nasdaq-100, with a 46% fall along the way against about 26% for either index.
  5. Currency conversion cost more than all commissions combined.
  6. A bot following the same rules mechanically would have done somewhat better, about 34% a year in a backtest. Adding a 15% stop-loss to that bot roughly halves its result.
  7. The period helped. Large US technology stocks, and a fast recovery from a crash, were the best possible conditions for buying drops.

The numbers

Measure Value
Net gain 75% on the money contributed
Trades 244 (146 buys, 98 sells)
Closed round trips 64, of which 60 made money
Median hold 35 days
Median sale 6% above average cost

I have left out the dollar size of the deposits, withdrawals and trades. Every figure below is a percentage, a count or a ratio.

How it unfolded

The account started in 2017 as a small portfolio of five large technology names bought on one day in January and held for months. By autumn I had started trading some of them quickly. It grew into a swing-trading account, and I emptied it in January 2021.

Year Trades Median hold Share of all realised gains
2017 28 92 days 3%
2018 57 25 days 7%
2019 58 43 days 12%
2020 90 31 days 59%
2021 11 63 days 19%

Realised gain is before commissions and counts in the year of the sale. Lot sizes roughly tripled between 2017 and 2018.

2017: a starter portfolio. Large names, bought near their highs, held for months. Only 20% of 2017 buys followed a sharp drop. The habit came later.

2018: the pattern sets in. Lot sizes roughly tripled and holding periods fell to about a month. A large transfer into the account arrived in late November, during the market sell-off, and went almost entirely into Amazon, Facebook, Nvidia and Alphabet within a week.

2019: the hardest year for the rules. Tesla and Nvidia kept falling through the spring after each add-on buy. Some lots were sold 17% down, and the Tesla position closed 6.8% below cost, the worst round trip in the record. Barnes & Noble Education, the first stock outside the well-known list, entered in March.

2020: the same rules meet a crash. The March sell-off and its recovery suited the pattern exactly. The small-cap was bought 19 times as it fell from $3.79 to $1.28 and sold in blocks at $1.94 to $2.46. Tesla and Nvidia were traded repeatedly on the way up. This one year produced 60% of all gains.

2021: the cash-out. Every position was sold in the first week of January, and the money was withdrawn over the following weeks.

The strategy as intended, and as traded

I had two rules in mind: only trade well-known companies, and buy when a sudden dip happens. Both held for most trades, and each has one clear exception.

Intended rule How it was tested Result
Only well-known companies Share of trades in large, familiar names 80% of trades, 77% of dollars bought
Buy on a sudden dip Stock down 3% or more that day, or 5% or more over the prior five trading days 58% of buys, against 16% of ordinary days

Well-known companies. Twenty of the 22 tickers fit. The exceptions are Barnes & Noble Education and ElectraMeccanica, both small speculative stocks. They account for 20% of trades and 38% of realised gains.

Sudden dips. The dip rule was not there at the start. The share of buys that followed a sharp drop was 20% in 2017, then 58% in 2018, 56% in 2019 and 69% in 2020. About a third of buys came in a slower slide, usually as add-ons to a position I already held.

What the memory leaves out. The description covers the entry only. The record shows three more habits that shaped the results as much: selling the rebound quickly, adding to falling positions, and refusing to sell at a loss. Over half of all buys (73 of the 136 that could be charted) were additions to a falling position, not first entries.

The eight rules, read back from the trades

These are what the record implies. They were not targets written down at the time.

  1. Universe. A short list of large US growth names traded again and again, plus one speculative small-cap at a time. Nvidia, Tesla and the small-cap are 48% of all trades.
  2. Open. Buy a round lot after a drop. 82% of buys were multiples of five shares; buys were larger and steadier from 2018 on.
  3. Add. If the price keeps falling, buy again. 81% of add-ons were below average cost, typically 3% under the previous buy and about a week later.
  4. Take profit. Sell on the rebound. The median sale was 6% above average cost; half fell between 2% and 13%.
  5. Exit size. Sell the whole position 65% of the time, otherwise usually half.
  6. Losses. No stop-loss. Only 9 of 98 sales were below average cost, and 4 of 64 round trips ended negative.
  7. Recycle. Put the proceeds straight back to work. Buys and sells fell on the same day on 23 of 153 trading days.
  8. Cash out. After a strong run, sell and withdraw.

The account was concentrated throughout. It usually held two to four positions, and the largest was typically 57% of the money deployed.

What the price charts show

Placed on daily candles, the trades line up with the stock’s range over the previous 20 trading days: buys cluster near the bottom of the range and sales near the top.

Measure At buys At sales Ordinary day
Price against the prior 20-day high (median) 12% below 2% below 4% below
Place in the 20-day range, 0% low to 100% high (median) 27% 83% 65%
Followed a falling week (buys) or rising week (sales) 72% 83% 41% / 58%
Below (buys) or above (sales) the 20-day average 74% 87% 38% / 62%
14-day RSI (median) 45 57 55
  • Price was the trigger, not an indicator. RSI, a common overbought and oversold gauge, was unremarkable at both ends. Only 8% of buys came with RSI under 30.
  • First buys were dips in an uptrend. They were typically 8% off the 20-day high but still above the 50-day and 200-day averages.
  • Add-on buys were made in downtrends. They were typically 16% off the 20-day high and 14% below the 200-day average.
  • Sales came on strength. Three in four were made on a day the stock closed up, and 54% were within 3% of the 20-day high.

Where the gains came from

The strategy won often and in small amounts: 81% of matched lots and 60 of 64 round trips closed at a gain, with a median gain of 6% per lot. Two names carried the total.

Tesla and Barnes & Noble Education each produced about a third of the realised gains, together about two-thirds. They earned them differently. Tesla’s came from 39 trades with a median lot gain of 6%, repeated many times with large positions. The small-cap’s came from a deep averaging-down campaign with a median lot gain of 21%. Amazon was the only ticker to lose money, after a five-buy position opened in October 2018 closed in March 2019 at 1.5% below cost.

By year, the gains followed the market more than the trade count. 2019 had 58 trades and 12% of the gains, with 69% of lots profitable. 2020 had 90 trades and 59%.

The gain in Canadian dollars

Measured in Canadian dollars, the account returned 75% on the money contributed, or about 35% a year when weighted by the timing of each deposit and withdrawal. Here is where that 75% came from, as a share of the money put in.

Component % of money contributed
Trading gains, each buy and sale converted at that day’s rate +83.2
Dividends +0.2
Commissions −2.8
Questrade’s currency conversion spread −4.6
Other: wire fees, currency ETF, drift on idle US cash −0.4
Net gain +75.5

The net gain is exact. The lines above it are approximate, because they rely on daily market exchange rates.

Currency conversion was the largest cost. Across 23 conversions, the broker’s rate was typically 1.6% worse than the market rate. That cost about 4.6% of the money contributed, more than all commissions combined. The conversion back to Canadian dollars in January 2021 alone cost about 2% of it.

The exchange rate worked slightly against the account. The same trading gains converted only at sale-date rates would be about 5 percentage points higher. The difference is the Canadian dollar strengthening while US positions were held, mostly in late 2020 and January 2021.

Measured like a fund

Treated as a fund, the account grew 27.7% a year in Canadian dollars from 6 January 2017 to 27 January 2021, against 14.1% for the S&P 500 and 26.5% for the Nasdaq-100. This measure ignores the timing and size of deposits and withdrawals, which is how fund returns are reported. C$10,000 at inception became C$26,946.

1 year 3 years, per year Since inception, per year Cumulative Volatility Worst drawdown
This account +73.9% +33.1% +27.7% +169.5% 38.7% −45.8%
S&P 500 +12.3% +12.3% +14.1% +70.5% 19.7% −27.7%
Nasdaq-100 +39.7% +25.1% +26.5% +159.4% 23.0% −25.6%

The account beat both indexes in 2018 and 2020, the two years with sharp sell-offs, where dip-buying had the most to work with. It trailed both in 2017, mainly because a block of AMD shares bought in October was below cost at year-end.

The return came with a rougher ride than either index. The account fell 46% between 4 March and 3 April 2020, despite a deposit during the fall, and recovered by late May. Its worst month was −33.8% in March 2020 and its best was +55.6% in April 2020.

Three more points:

  • Against the same cash flows, the gap is wider. The same deposits and withdrawals put into index funds would have gained about 25% of the money contributed in the S&P 500 or 56% in the Nasdaq-100, against 75% for the account.
  • The account was about two-thirds invested on average. It earned that return with less market exposure than the indexes had, the rest sitting in cash between trades.
  • Two return figures, two questions. 27.7% a year is the fund-style return for comparing against an index. The 35% a year in the previous section weights each dollar by when it was in the account, and is higher because most of the money was there during 2020.

If a bot had run the rules

A bot following the inferred rules mechanically, on the same stocks and with the same deposits and withdrawals, would have returned about 96% of the money contributed against the actual 75%, or 34.1% a year against 27.7%. The result moves a great deal with the settings, so treat it as a range, not a point.

Rule Bot setting
Universe The same stocks, each tradable only from the date the account first traded it
Buy After a day down 3% or more, or five days down 5% or more, buy at the next open with 10% of the account
Add Another 10% when the price is 3% below the last buy and five trading days have passed, up to 50% of the account in one stock
Sell The whole position at 6% above average cost
Losses No stop-loss
Costs and cash US$4.95 a trade, the same 1.6% currency conversion cost, the same deposits and withdrawals

How it differed from the real account. It traded more (554 trades against 244) and held for a median of 15 days against 35. It never chose to sell at a loss: its only losing exits were forced, three for the 2019 withdrawal and one at the final liquidation. It fell less, 36% at worst against 46%. The real account won in 2018 and 2020, where the large discretionary bets outdid the bot’s even 10% tranches; the bot was far ahead in 2019, 63% against 28%.

Change to the bot Gain, % of money contributed
Base case 96
Profit target 4% 98
Profit target 10% 124
Profit target 15% 121
Profit target 25% 89
Tranche 5% of account 59
Tranche 20% of account 75
Without the small-cap 113
With a 15% stop-loss 49

A stop-loss roughly halves the result, and a 10% to 15% profit target does better than 6%. Both fit the earlier findings: the exits were early, and the approach depends on stocks recovering.

Why this flatters the bot. The stock list was chosen by a person with knowledge of the time, and every name recovered. The rules were fitted to the same period they are tested on, in a strong market for these stocks. Two thinly documented stocks are left out for lack of price data, and fills are assumed at the open and at the target price with no slippage. The backtest shows that the rules were workable as a mechanical system in this period. It does not show they would work in another.

Where the strategy was exposed

The record shows four places where the outcome depended on conditions the rules did not control. These are observations from the trades, not advice.

Averaging down with no exit rule. After 60% of buys, the price fell more than another 5% within 20 trading days. The response was always to add, never to cut. This worked because every name eventually recovered within the holding window. The small-cap in 2020 is the stress case: by early April the oldest lots were sold 27% to 39% below their purchase prices, and the largest sale only broke even because of a big block bought near $1.60 in March.

Selling early in a rising market. After 65% of sales, the stock was higher 60 trading days later. The median stock sold was 86% higher by the end of March 2021. The 6% habit kept the win rate high and gave up most of the long-run rise in the same names being traded.

Concentration. The largest position was typically 57% of the money deployed, and at times one stock was most of the account. In March 2019 a large block of Tesla was bought in three weeks.

The speculative side. The two small-caps outside the stated universe took 23% of dollars bought and returned 38% of the gains. A small-cap that did not recover would have met a rule set with no stop-loss.

The market helped. 2017 to early 2021 was a strong period for large US technology stocks, including a fast recovery from the March 2020 crash. A dip-buying, no-stop-loss approach has not been tested here against a long decline.

What I take from it

  • Write the rules down, including the exit. The strategy I remembered covered only the entry. The behaviour that decided the outcome, selling early, adding to losers and never cutting, was not part of the story I told myself.
  • Compare against an index, in your own currency, after every cost. The account beat the S&P 500 by a wide margin and roughly matched the Nasdaq-100 on a fund-style basis (27.7% against 26.5% a year), with about 1.7 times the volatility and a fall of nearly half against a quarter. Most of the extra risk bought very little extra return over simply holding the Nasdaq-100.
  • Count the cost nobody shows you. Currency conversion cost more than commissions. A broker’s exchange rate is a fee even when the statement calls the trade free.
  • Ask which year did the work. One crash and its recovery produced most of the gains. A record that depends on one year says little about the other four.
  • Do not read a 60-of-64 win rate as skill. Small, frequent wins with no loss-cutting look excellent until the one position that does not recover. This one never met it, which is not the same as being safe.

Method and data limits

  • Source. The account’s 291 activity rows, spot-checked against the broker’s live records.
  • Gains. Sales were matched to purchases first-in, first-out, giving 162 lot matches. Gains by ticker and by year are sale price less purchase price, before commissions. The after-commission total is the net cash from all trades.
  • Prices. Daily candles from Yahoo Finance, converted back to the share prices of the time. All 228 charted trades fall inside that day’s high-to-low range.
  • Not charted. Two stocks (16 trades) have no price history available from Yahoo or the broker. They are included in gains and counts but not in the dip and candle figures.
  • Not covered. A retirement account that traded some of the same names at the same time, and any account activity after March 2021.
  • Currency. Gains are mostly in US dollars and cash flows in Canadian dollars. The net gain in Canadian dollars is exact, from the account’s own cash flows. The breakdown of that gain uses daily market rates and is approximate.
  • Interpretation. The figures show when trades happened relative to price, not why. Daily data cannot show whether a buy came from a limit order, a news event or an intraday move.
  • Fund-style returns. Daily account values built from cash in both currencies plus positions at closing prices. Returns are time-weighted.
  • Benchmarks. The S&P 500 and Nasdaq-100 are the SPY and QQQ funds with dividends reinvested, converted to Canadian dollars at the same daily rates.
  • Bot. A day-by-day simulation on split-adjusted daily prices. It is a backtest, not a record of trades.

A second test: January 2020 to December 2024

Would buying the drop have beaten simply holding the same stocks? To separate stock selection from the timing rule, I tested both approaches on two lists. These are simulated results, separate from the actual account record above.

The first list was Microsoft (MSFT), Alphabet (GOOGL), NVIDIA (NVDA), Shopify’s Toronto listing (SHOP.TO) and Netflix (NFLX). The second was Tesla (TSLA) and NIO (NIO). Both lists were chosen after the fact. Barnes & Noble Education was excluded because its 2024 rights offering and reverse split cannot be represented fairly by split-adjusted prices alone.

Stocks Approach Annualized return (CAD) Maximum drawdown Buy and sell fills
Starting five Buy the drop with adds 24.1% −61.5% 287
Starting five Buy the drop, no adds 12.1% −25.7% 179
Starting five Equal-weight at entry, then hold 49.1% −58.7% 5
TSLA and NIO Buy the drop with adds 2.6% −71.5% 171
TSLA and NIO Buy the drop, no adds 6.8% −13.6% 80
TSLA and NIO Equal-weight at entry, then hold 53.3% −74.1% 2
SPY Buy and hold 16.3% −27.4% 1
QQQ Buy and hold 21.8% −32.0% 1

On the first list, the dip strategy beat QQQ by 2.3 percentage points a year, but its drawdown was almost twice as deep. Holding the same five did far better on return, helped by staying invested in NVIDIA through its rise. Without NVIDIA, the dip strategy returned 14.1% annualized, below both ETFs. The no-add version fell less, but it also spent much less of the portfolio invested.

How the test was run

Each run began with CA$10,000, with no later deposits, withdrawals or leverage. A drop of at least 3% in one regular session or 5% over five sessions triggered a buy at the next available regular-session open. The first buy used up to 10% of equity. The add version allowed another purchase of up to 10% at least five sessions after the previous fill if the close was at least 3% below that fill; the position was capped at 50% at purchase. The no-add version capped the initial purchase at 10%. Both versions signaled a full exit when the close reached 6% above average purchase cost, then sold at the next open. Neither used a stop-loss, and the next-open exit did not guarantee a 6% fill.

The simulation charged 4.95 in the instrument’s currency per fill, 0.10% one-way slippage and 1.6% on currency conversions. The main run reused USD proceeds in USD. Converting after every USD sale reduced the starting-five dip return from 24.1% to 14.5% annualized. Buy-and-hold portfolios reinvested ordinary distributions; the dip strategies retained them as cash. Open positions were marked at the end of 2024, not sold. Taxes, settlement delays and market impact were not modeled.

Daily bars came from Nasdaq historical data for U.S. listings and Yahoo Finance for SHOP.TO. Distributions were checked against Nasdaq or fund-provider records. USD values were translated using Bank of Canada daily rates, using the last available rate for next-open conversions.

This is a sensitivity check, not independent evidence that the rule works. Early 2020 through March 2021 overlaps the original sample, both stock lists were picked retrospectively, and the test did not include a historical universe with delisted names. A five-year test window is not a five-year holding rule. The test did not use a forward valuation filter; that would require estimates available at the time of each signal. The takeaway is to treat a drop as a reason to investigate, not as a buy signal by itself.

This is a description of one person’s account in one period, not investment, tax or legal advice. Past results, including a backtest of them, do not predict future ones.

© 2026 Collegica

A learning companion to Mechatronics3D

  • About

  • Apps

  • Events

  • Talks

  • YouTube

  • Facebook

  • For AI agents