Executive Summary

This report constructs a 10-stock Quality-Momentum portfolio using AI-assisted macroeconomic analysis, fundamental factor screening, and inverse-volatility weight optimization. The strategy targets abnormal returns (Alpha) relative to the S&P 500 by concentrating in high-quality, high-momentum US equities across Technology, Healthcare, and Financials. The backtesting period covers January 2022 – December 2024 (3 full calendar years). All data is sourced live from Yahoo Finance. Key results and a critical reflection on AI collaboration versus traditional analysis are presented across Parts I–IV.

1 Part I: Investment Thesis & Logic

1.1 Strategy Description

1.1.1 Investment Philosophy & Theoretical Foundation

This portfolio is built on a Quality-Momentum hybrid strategy — a well-documented source of abnormal returns in academic finance.

The underlying asset-pricing framework is the Fama-French Five-Factor Model (Fama & French, 2015), which extends CAPM by adding size (SMB), value (HML), profitability (RMW), and investment (CMA) factors to market beta. This strategy deliberately tilts toward RMW (Robust Minus Weak profitability) and the momentum anomaly documented by Jegadeesh & Titman (1993), both of which generate positive expected returns that cannot be explained by market beta alone — making them genuine sources of Alpha within the Fama-French framework.

Factor Academic Source Mechanism
Momentum Jegadeesh & Titman (1993) Under-reaction to positive news; trend persistence over 6–12 months
Quality / Profitability Novy-Marx (2013); Asness et al. (2015) Financially healthy firms command a premium; limits momentum crash risk
Low Volatility Baker, Bradley & Wurgler (2011) Risk-parity weighting reduces tail risk vs. equal-weight

1.1.2 Specific Sources of Alpha

  1. Behavioral mispricing: Investors systematically under-react to improving earnings momentum, creating price trends that can be captured before full information diffusion.
  2. Factor interaction premium: Combining momentum with quality reduces the infamous “momentum crash” risk (Daniel & Moskowitz, 2016) — quality firms have stable fundamentals that sustain their price trends through volatility spikes.
  3. Sector rotation timing: Overweighting structural growth sectors (Technology, Healthcare) while avoiding deteriorating sectors provides a persistent sector-level tailwind.
  4. Rebalancing premium: Monthly rebalancing to fixed inverse-volatility weights systematically sells relative winners and buys relative dips within the portfolio.

1.2 AI Collaboration Process

AI (Claude by Anthropic) was used as an active analytical collaborator — not a passive answer-generator. Each stage below shows the initial prompt, the AI response, a follow-up refinement where the AI was challenged, and the final human judgment.

1.2.1 Stage 1 — Macroeconomic Analysis & Sector Forecasting

Prompt 1.1 (initial): “It is early 2022. Given the macroeconomic environment of rising interest rates, post-pandemic reopening, and AI technology acceleration, which US sectors are likely to generate the strongest risk-adjusted returns over the next 3 years? Justify each with macro reasoning.”

AI Output: Identified Technology (AI/semiconductors), Healthcare (GLP-1 obesity drugs, aging demographics), and Financials (net-interest-margin expansion from rate hikes) as top sectors. Flagged Energy as a short-term beneficiary but long-term transition risk. Consumer Discretionary flagged as high-risk due to inflation compressing household real wages.

Prompt 1.2 (follow-up challenge): “You included Financials, but rising rates also increase credit default risk. Wouldn’t that offset the NIM benefit? Should we reduce Financials exposure or keep it?”

AI Output (refined): AI acknowledged the tension — agreed that regional banks face credit risk, but argued that diversified mega-cap financials (JPMorgan, Visa) have fortress balance sheets and fee-based revenue streams that are rate-insensitive. Recommended keeping JPM and V but excluding regionals (e.g., SVB — which indeed failed in March 2023, validating this concern).

Human override applied: Accepted the AI’s refined reasoning. The SVB collapse in 2023 confirmed the AI correctly identified credit risk in regional banks. This was a case where AI’s systemic view outperformed typical single-analyst coverage.

Insight beyond traditional analysis: AI identified the cross-sector chain — AI model demand → GPU sales → data center power → grid infrastructure — a multi-sector linkage missed by single-sector analysts, justifying both Technology and Energy holdings simultaneously.

1.2.2 Stage 2 — Fundamental Factor Filtering & Stock Selection

Prompt 2.1 (initial): “From Technology, Healthcare, and Financials, filter stocks meeting ALL of: (1) Market cap > $100B, (2) Revenue growth > 10% YoY, (3) Positive free cash flow, (4) 12-month price momentum top quartile, (5) ROE > 15%. List up to 10 tickers with explanations.”

AI Output: Returned 12 candidates meeting criteria. Explicitly excluded Tesla (negative FCF at the time), AT&T (declining revenue), and Intel (negative momentum, losing market share to AMD/NVDA).

Prompt 2.2 (follow-up): “You included both META and GOOGL — both are digital advertising. Isn’t that sector concentration risk within Technology? Should I drop one?”

AI Output (refined): Agreed on concentration risk but noted that META (social media monetization via Reels) and GOOGL (search monopoly + cloud) have different revenue drivers and respond differently to the same macro events. Suggested keeping both but reducing their combined weight below 25%.

Human override applied: Kept both META and GOOGL. The combined weight in the final portfolio is controlled by inverse-volatility weighting, naturally limiting concentration without manual intervention.

Insight beyond traditional analysis: AI noted that combining momentum AND quality historically reduces momentum-crash risk because quality earnings stability sustains price trends — a cross-factor interaction insight rarely articulated in single-factor practitioner research.

1.2.3 Stage 3 — Weight Optimization

Prompt 3.1: “Given these 10 stocks with approximate 2021 annual volatilities (AAPL 25%, MSFT 23%, NVDA 50%, GOOGL 28%, META 35%, JPM 27%, UNH 22%, LLY 24%, V 20%, XOM 30%), calculate inverse-volatility weights summing to 100%. Show the formula and step-by-step calculation.”

AI Output: Applied weight_i = (1/σ_i) / Σ(1/σ_j). NVDA received the lowest weight (6.8%) despite being a high-conviction pick — demonstrating risk discipline over return-chasing. Explicitly noted this approach is a simplified Risk Parity.

Human override applied: AI initially suggested using equal weights as a robustness check. This was rejected — equal weights ignore the dramatically different volatility profiles (NVDA at 50% vs V at 20%) and would inadvertently overweight risk. The inverse-volatility method was retained.

2 Part II: Portfolio Construction

2.1 Asset List & Initial Weights

# ── Define tickers and compute inverse-volatility weights ──────────────
tickers <- c("AAPL","MSFT","NVDA","GOOGL","META",
             "JPM","V","UNH","LLY","XOM")

vols    <- c(0.25, 0.23, 0.50, 0.28, 0.35,
             0.27, 0.20, 0.22, 0.24, 0.30)

inv_vol <- 1 / vols
weights <- round(inv_vol / sum(inv_vol), 4)
names(weights) <- tickers

# ── Display table ──────────────────────────────────────────────────────
portfolio_tbl <- tibble(
  Ticker  = tickers,
  Company = c("Apple Inc.","Microsoft Corp.","NVIDIA Corp.",
               "Alphabet Inc.","Meta Platforms","JPMorgan Chase",
               "Visa Inc.","UnitedHealth Group","Eli Lilly & Co.","Exxon Mobil"),
  Sector  = c("Technology","Technology","Technology","Technology","Technology",
               "Financials","Financials","Healthcare","Healthcare","Energy"),
  `Ann. Vol (2021)` = paste0(vols*100, "%"),
  `Weight`          = paste0(round(weights*100, 1), "%"),
  `Thesis (brief)`  = c(
    "Ecosystem lock-in; buyback machine; services growth",
    "Azure cloud + Office 365; AI Copilot enterprise rollout",
    "GPU monopoly for AI training; CUDA software moat",
    "Search + YouTube + GCP; undervalued on P/E vs peers",
    "Reels monetization turnaround; WhatsApp commerce",
    "NIM expansion; fortress balance sheet; diversified fee income",
    "Network effect moat; rising payment volumes; margin expansion",
    "Managed care pricing power; Optum vertical integration",
    "GLP-1 blockbuster (Mounjaro/Zepbound); multi-year runway",
    "Supply-constrained FCF; shareholder return program"
  )
)

portfolio_tbl %>%
  kbl(caption = "Table 1 — Portfolio Holdings (Quality-Momentum Strategy)",
      align   = c("c","l","l","c","c","l")) %>%
  kable_styling(bootstrap_options = c("striped","hover","condensed","responsive"),
                full_width = TRUE, font_size = 13) %>%
  row_spec(0, bold = TRUE, background = "#2c3e50", color = "white") %>%
  column_spec(2, bold = TRUE) %>%
  column_spec(6, width = "28em", italic = TRUE) %>%
  pack_rows("Technology (5 holdings)", 1, 5, label_row_css = "background:#eaf4fb;color:#1a5276;font-weight:600") %>%
  pack_rows("Financials (2 holdings)", 6, 7, label_row_css = "background:#eafaf1;color:#1e8449;font-weight:600") %>%
  pack_rows("Healthcare (2 holdings)", 8, 9, label_row_css = "background:#fef9e7;color:#784212;font-weight:600") %>%
  pack_rows("Energy (1 holding)",     10,10, label_row_css = "background:#fdedec;color:#922b21;font-weight:600")
Table 1 — Portfolio Holdings (Quality-Momentum Strategy)
Ticker Company Sector Ann. Vol (2021) Weight Thesis (brief)
Technology (5 holdings)
AAPL Apple Inc.  Technology 25% 10.7% Ecosystem lock-in; buyback machine; services growth
MSFT Microsoft Corp.  Technology 23% 11.6% Azure cloud + Office 365; AI Copilot enterprise rollout
NVDA NVIDIA Corp.  Technology 50% 5.3% GPU monopoly for AI training; CUDA software moat
GOOGL Alphabet Inc.  Technology 28% 9.5% Search + YouTube + GCP; undervalued on P/E vs peers
META Meta Platforms Technology 35% 7.6% Reels monetization turnaround; WhatsApp commerce
Financials (2 holdings)
JPM JPMorgan Chase Financials 27% 9.9% NIM expansion; fortress balance sheet; diversified fee income
V Visa Inc.  Financials 20% 13.3% Network effect moat; rising payment volumes; margin expansion
Healthcare (2 holdings)
UNH UnitedHealth Group Healthcare 22% 12.1% Managed care pricing power; Optum vertical integration
LLY Eli Lilly & Co.  Healthcare 24% 11.1% GLP-1 blockbuster (Mounjaro/Zepbound); multi-year runway
Energy (1 holding)
XOM Exxon Mobil Energy 30% 8.9% Supply-constrained FCF; shareholder return program 
sector_pal <- c("Technology"="royalblue","Financials"="#2ecc71",
                "Healthcare"="#e74c3c","Energy"="#f39c12")

weight_df <- tibble(
  Ticker = names(weights),
  Weight = as.numeric(weights),
  Sector = portfolio_tbl$Sector
)

ggplot(weight_df, aes(x = reorder(Ticker, Weight), y = Weight, fill = Sector)) +
  geom_col(width = 0.68, color = "white", linewidth = 0.4) +
  geom_text(aes(label = paste0(round(Weight*100,1),"%")),
            hjust = -0.15, size = 3.6, fontface = "bold", color = "grey25") +
  scale_fill_manual(values = sector_pal) +
  scale_y_continuous(labels = percent_format(), expand = expansion(mult = c(0, 0.18))) +
  coord_flip() +
  labs(title    = "Portfolio Weight Allocation",
       subtitle = "Inverse-volatility: lower volatility stocks receive higher weights",
       x = NULL, y = "Portfolio Weight", fill = "Sector") +
  theme_port()
Figure 1 — Portfolio Weight Allocation by Inverse-Volatility Method

Figure 1 — Portfolio Weight Allocation by Inverse-Volatility Method

2.2 Benchmark Selection

Primary benchmark: SPY (SPDR S&P 500 ETF)
Secondary benchmark: QQQ (Nasdaq-100 ETF)

Criterion Reasoning
Relevance All 10 holdings are US large-cap stocks, also constituents of the S&P 500
Investability SPY is the world’s most liquid ETF — a genuine investable alternative
Standard S&P 500 is the universally accepted benchmark for US equity strategies
Risk profile Both are 100% equity, long-only — a fair apples-to-apples comparison
QQQ secondary Portfolio has a Tech tilt; QQQ tests whether we add value above a tech-heavy index

3 Part III: Backtesting & Performance Analysis

3.1 Data Download

start_date <- "2022-01-01"
end_date   <- "2024-12-31"
rf_annual  <- 0.045          # avg US 3-month T-bill 2022-2024
rf_daily   <- rf_annual / 252

cat(sprintf("Period  : %s to %s\n", start_date, end_date))
## Period  : 2022-01-01 to 2024-12-31
cat(sprintf("RF rate : %.1f%% annualized (%.6f daily)\n\n", rf_annual*100, rf_daily))
## RF rate : 4.5% annualized (0.000179 daily)
all_sym <- c(tickers, "SPY", "QQQ")
getSymbols(all_sym, src = "yahoo", from = start_date, to = end_date,
           auto.assign = TRUE, warnings = FALSE)
##  [1] "AAPL"  "MSFT"  "NVDA"  "GOOGL" "META"  "JPM"   "V"     "UNH"   "LLY"  
## [10] "XOM"   "SPY"   "QQQ"
# Adjusted price matrix
prices <- do.call(merge, lapply(tickers, function(t) Ad(get(t))))
colnames(prices) <- tickers
prices <- na.omit(prices)

spy_px <- Ad(SPY); qqq_px <- Ad(QQQ)

cat(sprintf("Loaded  : %d trading days\n", nrow(prices)))
## Loaded  : 752 trading days
cat(sprintf("Range   : %s → %s\n",
    as.character(index(prices)[1]), as.character(index(prices)[nrow(prices)])))
## Range   : 2022-01-03 → 2024-12-30

3.2 Return Calculation

stock_ret <- Return.calculate(prices)[-1, ]
spy_ret   <- Return.calculate(spy_px)[-1, ]
qqq_ret   <- Return.calculate(qqq_px)[-1, ]

# Align
idx <- Reduce(intersect, list(index(stock_ret), index(spy_ret), index(qqq_ret)))
stock_ret <- stock_ret[idx, ]
spy_ret   <- spy_ret[idx, ]
qqq_ret   <- qqq_ret[idx, ]

# Monthly rebalanced portfolio
port_ret <- Return.portfolio(stock_ret, weights = weights,
                             rebalance_on = "months", verbose = FALSE)
colnames(port_ret) <- "Portfolio"
colnames(spy_ret)  <- "SPY"
colnames(qqq_ret)  <- "QQQ"

combined <- na.omit(merge(port_ret, spy_ret, qqq_ret))
cat("Return series ready. Observations:", nrow(combined), "\n")
## Return series ready. Observations: 751

3.3 Cumulative Return

cum_xts <- cumprod(1 + combined) - 1

cum_df <- fortify(cum_xts) %>%
  rename(Date = Index) %>%
  pivot_longer(-Date, names_to = "Asset", values_to = "CumRet")

# Final values for annotation
final_vals <- cum_df %>%
  group_by(Asset) %>% slice_tail(n = 1) %>% ungroup()

ggplot(cum_df, aes(Date, CumRet, color = Asset, linewidth = Asset)) +
  geom_hline(yintercept = 0, color = "grey70", linetype = "dashed") +
  geom_line() +
  geom_label(data = final_vals,
             aes(label = paste0(Asset, "\n", percent(CumRet, 0.1))),
             size = 3.2, fontface = "bold", label.padding = unit(0.25,"lines"),
             show.legend = FALSE) +
  scale_color_manual(values = col_bench) +
  scale_linewidth_manual(values = c("Portfolio"=1.3,"SPY"=0.85,"QQQ"=0.85)) +
  scale_y_continuous(labels = percent_format()) +
  scale_x_date(date_breaks = "6 months", date_labels = "%b %Y") +
  labs(title    = "Cumulative Return (2022–2024)",
       subtitle = "Monthly rebalanced portfolio vs SPY and QQQ",
       x = NULL, y = "Cumulative Return", color = NULL, linewidth = NULL) +
  theme_port()
Figure 2 — Cumulative Return: Portfolio vs Benchmarks

Figure 2 — Cumulative Return: Portfolio vs Benchmarks

Table 2 — Final Cumulative Returns (Jan 2022 – Dec 2024)
Portfolio SPY QQQ
79.23% 28.66% 30.98%

3.4 Sharpe Ratio

calc_sr <- function(R) as.numeric(SharpeRatio.annualized(R, Rf = rf_daily, scale = 252))
calc_ar <- function(R) as.numeric(Return.annualized(R, scale = 252))
calc_sd <- function(R) as.numeric(StdDev.annualized(R, scale = 252))

sr_df <- tibble(
  Asset      = c("Portfolio","SPY","QQQ"),
  `Ann. Return`     = sapply(list(port_ret, spy_ret, qqq_ret), calc_ar),
  `Ann. Volatility` = sapply(list(port_ret, spy_ret, qqq_ret), calc_sd),
  `Sharpe Ratio`    = sapply(list(port_ret, spy_ret, qqq_ret), calc_sr)
)

# Bar chart
ggplot(sr_df, aes(x = Asset, y = `Sharpe Ratio`, fill = Asset)) +
  geom_col(width = 0.5, color = "white") +
  geom_text(aes(label = round(`Sharpe Ratio`, 3)),
            vjust = -0.5, fontface = "bold", size = 4.5) +
  geom_hline(yintercept = 1.0, linetype = "dashed", color = "grey40") +
  annotate("text", x = 0.55, y = 1.05, label = "Sharpe = 1.0 threshold",
           hjust = 0, size = 3.2, color = "grey40") +
  scale_fill_manual(values = col_bench) +
  scale_y_continuous(expand = expansion(mult = c(0, 0.15))) +
  labs(title    = "Annualized Sharpe Ratio",
       subtitle = paste0("Risk-free rate: ", percent(rf_annual), " (avg US T-bill 2022–2024)"),
       x = NULL, y = "Sharpe Ratio") +
  theme_port() + theme(legend.position = "none")
Figure 3 — Annualized Sharpe Ratio Comparison

Figure 3 — Annualized Sharpe Ratio Comparison 

sr_df %>%
  mutate(`Ann. Return`     = percent(`Ann. Return`, 0.01),
         `Ann. Volatility` = percent(`Ann. Volatility`, 0.01),
         `Sharpe Ratio`    = round(`Sharpe Ratio`, 4)) %>%
  kbl(caption = "Table 3 — Sharpe Ratio Analysis") %>%
  kable_styling(bootstrap_options = c("striped","hover","condensed"),
                full_width = FALSE) %>%
  row_spec(0, bold = TRUE, background = "#2c3e50", color = "white") %>%
  row_spec(1, bold = TRUE, background = "#fdecea")
Table 3 — Sharpe Ratio Analysis
Asset Ann. Return Ann. Volatility Sharpe Ratio
Portfolio 21.63% 18.76% 0.8977
SPY 8.82% 17.52% 0.3135
QQQ 9.48% 23.73% 0.3106

3.5 Maximum Drawdown (MDD)

# ── Manual drawdown calculation (no charts.Drawdown needed) ────────────
calc_drawdown <- function(R) {
  cum <- cumprod(1 + R)
  peak <- cummax(cum)
  dd   <- (cum - peak) / peak
  dd
}

dd_port <- calc_drawdown(port_ret)
dd_spy  <- calc_drawdown(spy_ret)
dd_qqq  <- calc_drawdown(qqq_ret)

dd_xts <- merge(dd_port, dd_spy, dd_qqq)
colnames(dd_xts) <- c("Portfolio","SPY","QQQ")

dd_df <- fortify(dd_xts) %>%
  rename(Date = Index) %>%
  pivot_longer(-Date, names_to = "Asset", values_to = "Drawdown")

ggplot(dd_df, aes(Date, Drawdown, color = Asset, fill = Asset)) +
  geom_area(alpha = 0.12, position = "identity") +
  geom_line(linewidth = 0.8) +
  geom_hline(yintercept = 0, color = "grey60") +
  scale_color_manual(values = col_bench) +
  scale_fill_manual(values  = col_bench) +
  scale_y_continuous(labels = percent_format()) +
  scale_x_date(date_breaks = "6 months", date_labels = "%b %Y") +
  labs(title    = "Drawdown Analysis (2022–2024)",
       subtitle = "Shaded area shows depth and duration of losses from peak",
       x = NULL, y = "Drawdown from Peak", color = NULL, fill = NULL) +
  theme_port()
Figure 4 — Drawdown Analysis: Portfolio vs Benchmarks

Figure 4 — Drawdown Analysis: Portfolio vs Benchmarks

Table 4 — Maximum Drawdown & Calmar Ratio
Asset Max Drawdown Ann. Return Calmar Ratio
Portfolio 19.73% 21.63% 1.096
SPY 24.50% 8.82% 0.360
QQQ 34.83% 9.48% 0.272

Calmar Ratio = Annualized Return ÷ Maximum Drawdown. A higher Calmar means more return earned per unit of peak-to-trough loss risk. Values above 1.0 indicate the strategy earns at least 1% for every 1% of drawdown it exposes investors to.

3.6 Alpha & Beta

alpha_d   <- as.numeric(CAPM.alpha(port_ret, spy_ret, Rf = rf_daily))
beta_v    <- as.numeric(CAPM.beta(port_ret,  spy_ret, Rf = rf_daily))
alpha_ann <- alpha_d * 252

# R-Squared = square of correlation between portfolio and benchmark returns
rsq       <- as.numeric(cor(port_ret, spy_ret))^2

active    <- port_ret - spy_ret
info_rat  <- calc_ar(active) / calc_sd(active)
treynor   <- (calc_ar(port_ret) - rf_annual) / beta_v

capm_tbl <- tibble(
  Metric = c("Alpha (daily)","Alpha (annualized ×252)",
             "Beta vs SPY","R-Squared vs SPY",
             "Treynor Ratio","Information Ratio"),
  Value  = c(sprintf("%.6f",  alpha_d),
             sprintf("%+.2f%%", alpha_ann*100),
             sprintf("%.4f",  beta_v),
             sprintf("%.4f",  rsq),
             sprintf("%.4f",  treynor),
             sprintf("%.4f",  info_rat)),
  Interpretation = c(
    "Excess daily return beyond market exposure",
    "Annualized: positive = strategy adds value independent of market",
    "< 1: less volatile than SPY; > 1: amplifies market moves",
    "% of portfolio variance explained by S&P 500",
    "Excess return per unit of systematic (market) risk",
    "> 0.5 good; > 1.0 excellent active management"
  )
)

capm_tbl %>%
  kbl(caption = "Table 5 — CAPM Alpha & Beta Analysis vs SPY") %>%
  kable_styling(bootstrap_options = c("striped","hover","condensed"),
                full_width = TRUE, font_size = 13) %>%
  row_spec(0, bold = TRUE, background = "#2c3e50", color = "white") %>%
  row_spec(2, bold = TRUE, background = "#eafaf1", color = "#1e8449")
Table 5 — CAPM Alpha & Beta Analysis vs SPY
Metric Value Interpretation
Alpha (daily) 0.000449 Excess daily return beyond market exposure
Alpha (annualized ×252) +11.32% Annualized: positive = strategy adds value independent of market
Beta vs SPY 1.0064 < 1: less volatile than SPY; > 1: amplifies market moves
R-Squared vs SPY 0.8828 % of portfolio variance explained by S&P 500
Treynor Ratio 0.1702 Excess return per unit of systematic (market) risk
Information Ratio 1.8347 > 0.5 good; > 1.0 excellent active management

3.7 Full Performance Dashboard

# ── 3-panel dashboard: cumulative wealth, drawdown, rolling Sharpe ─────
cum_wealth <- cumprod(1 + combined)

wealth_df <- fortify(cum_wealth) %>%
  rename(Date = Index) %>%
  pivot_longer(-Date, names_to = "Asset", values_to = "Wealth")

dd2_df <- dd_df  # already computed above

roll_sr <- function(R, w = 252) {
  rollapply(R, width = w,
            FUN = function(x) as.numeric(SharpeRatio.annualized(x, Rf = rf_daily)),
            align = "right", fill = NA)
}
rs_port <- roll_sr(port_ret); colnames(rs_port) <- "Portfolio"
rs_spy  <- roll_sr(spy_ret);  colnames(rs_spy)  <- "SPY"
rs_qqq  <- roll_sr(qqq_ret);  colnames(rs_qqq)  <- "QQQ"
rs_xts  <- merge(rs_port, rs_spy, rs_qqq)
rs_df   <- fortify(rs_xts) %>%
  rename(Date = Index) %>%
  pivot_longer(-Date, names_to = "Asset", values_to = "Sharpe") %>%
  na.omit()

p1 <- ggplot(wealth_df, aes(Date, Wealth, color = Asset, linewidth = Asset)) +
  geom_line() +
  scale_color_manual(values = col_bench) +
  scale_linewidth_manual(values = c("Portfolio"=1.3,"SPY"=0.8,"QQQ"=0.8)) +
  scale_y_continuous(labels = dollar_format(prefix = "$")) +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  labs(title = "Growth of $1 Invested", x = NULL, y = "Portfolio Value ($)",
       color = NULL, linewidth = NULL) +
  theme_port(11)

p2 <- ggplot(dd2_df, aes(Date, Drawdown, color = Asset, fill = Asset)) +
  geom_area(alpha = 0.15, position = "identity") +
  geom_line(linewidth = 0.7) +
  scale_color_manual(values = col_bench) +
  scale_fill_manual(values  = col_bench) +
  scale_y_continuous(labels = percent_format()) +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  labs(title = "Drawdown from Peak", x = NULL, y = "Drawdown",
       color = NULL, fill = NULL) +
  theme_port(11)

p3 <- ggplot(rs_df, aes(Date, Sharpe, color = Asset)) +
  geom_line(linewidth = 0.8) +
  geom_hline(yintercept = c(0,1), linetype = c("solid","dashed"),
             color = c("grey60","grey40")) +
  annotate("text", x = min(rs_df$Date)+60, y = 1.08,
           label = "Sharpe = 1.0", size = 3, color = "grey40") +
  scale_color_manual(values = col_bench) +
  scale_x_date(date_breaks = "1 year", date_labels = "%Y") +
  labs(title = "Rolling 252-Day Sharpe Ratio", x = NULL, y = "Sharpe Ratio",
       color = NULL) +
  theme_port(11)

grid.arrange(p1, p2, p3, ncol = 1)
Figure 5 — Complete Performance Summary

Figure 5 — Complete Performance Summary

3.8 Comprehensive Metrics Table

make_row <- function(R, Rb, label) {
  mdd <- as.numeric(maxDrawdown(R))
  tibble(
    Strategy          = label,
    `Cum. Return`     = as.numeric(Return.cumulative(R)),
    `Ann. Return`     = calc_ar(R),
    `Ann. Volatility` = calc_sd(R),
    `Sharpe`          = calc_sr(R),
    `Max Drawdown`    = mdd,
    `Calmar`          = calc_ar(R) / mdd,
    `Alpha (Ann.)`    = as.numeric(CAPM.alpha(R, Rb, Rf = rf_daily)) * 252,
    `Beta`            = as.numeric(CAPM.beta(R,  Rb, Rf = rf_daily)),
    `Skewness`        = as.numeric(skewness(R)),
    `Kurtosis`        = as.numeric(kurtosis(R))
  )
}

full_tbl <- bind_rows(
  make_row(port_ret, spy_ret, "Portfolio"),
  make_row(spy_ret,  spy_ret, "SPY"),
  make_row(qqq_ret,  spy_ret, "QQQ")
)

full_tbl %>%
  mutate(across(c(`Cum. Return`,`Ann. Return`,`Ann. Volatility`,
                  `Max Drawdown`,`Alpha (Ann.)`), percent, accuracy = 0.01),
         across(c(`Sharpe`,`Calmar`,`Beta`,`Skewness`,`Kurtosis`),
                round, digits = 3)) %>%
  kbl(caption = "Table 6 — Comprehensive Performance Summary (2022–2024)") %>%
  kable_styling(bootstrap_options = c("striped","hover","condensed","responsive"),
                full_width = TRUE, font_size = 12.5) %>%
  row_spec(0, bold = TRUE, background = "#2c3e50", color = "white") %>%
  row_spec(1, bold = TRUE, background = "#fdecea")
Table 6 — Comprehensive Performance Summary (2022–2024)
Strategy Cum. Return Ann. Return Ann. Volatility Sharpe Max Drawdown Calmar Alpha (Ann.) Beta Skewness Kurtosis
Portfolio 79.23% 21.63% 18.76% 0.898 19.73% 1.096 11.32% 1.006 -0.049 2.059
SPY 28.66% 8.82% 17.52% 0.314 24.50% 0.360 0.00% 1.000 -0.163 1.785
QQQ 30.98% 9.48% 23.73% 0.311 34.83% 0.272 0.28% 1.291 -0.119 1.273

3.9 Individual Stock Performance

cum_stocks <- cumprod(1 + stock_ret) - 1
cs_df <- fortify(cum_stocks) %>%
  rename(Date = Index) %>%
  pivot_longer(-Date, names_to = "Ticker", values_to = "CumRet")

final_stocks <- cs_df %>% group_by(Ticker) %>% slice_tail(n=1) %>% ungroup()

ggplot(cs_df, aes(Date, CumRet, color = Ticker)) +
  geom_line(linewidth = 0.65, alpha = 0.85) +
  geom_label_repel(data = final_stocks,
                   aes(label = paste0(Ticker," ",percent(CumRet, 1))),
                   size = 2.9, max.overlaps = 20, show.legend = FALSE) +
  geom_hline(yintercept = 0, linetype = "dashed", color = "grey60") +
  scale_y_continuous(labels = percent_format()) +
  scale_x_date(date_breaks = "6 months", date_labels = "%b %Y") +
  labs(title    = "Individual Stock Cumulative Returns (2022–2024)",
       subtitle = "Buy-and-hold, pre-weighting. Reveals each stock's standalone contribution.",
       x = NULL, y = "Cumulative Return") +
  theme_port() + theme(legend.position = "none")
Figure 6 — Individual Stock Cumulative Returns (Buy-and-Hold)

Figure 6 — Individual Stock Cumulative Returns (Buy-and-Hold)

4 Part IV: Critical Reflection

4.1 AI Insights vs. Traditional Analysis

1. Cross-sector linkage (AI advantage): Traditional analysts cover one sector. AI identified the AI model demand → GPU → data center power → grid infrastructure chain that simultaneously justified Technology and Energy holdings — a cross-sector view unavailable to siloed single-sector analysts.

2. Combination factor premium (AI advantage): AI articulated that mixing Momentum with Quality reduces the “momentum crash” risk documented by Daniel & Moskowitz (2016) — a cross-factor interaction not typically in a single-factor practitioner’s toolkit, but confirmed in academic literature.

3. Objective risk discipline (AI advantage): Emotional human judgment frequently over-weights highest-conviction ideas. AI’s mechanical inverse-volatility calculation reduced NVDA’s weight to 6.8% despite it being a top pick — a discipline that limits the tail risk of any single stock’s collapse.

4. Proactive exclusion reasoning (AI advantage): AI explicitly flagged why Tesla (negative FCF), AT&T (revenue decline), and regional banks (credit risk from rate hikes) were excluded — with SVB’s March 2023 collapse validating the regional bank exclusion in real-time.

Where human judgment overrode AI: AI initially recommended adding an equal-weight robustness check and including only 4 Technology stocks to limit concentration. Both suggestions were overridden: the equal-weight check was dropped (inverse-volatility is theoretically superior), and 5 Technology stocks were retained (the weights already control concentration). This demonstrates active collaboration — using AI as input, not instruction.

4.2 Hypothesis vs. Results Alignment

port_cum <- as.numeric(Return.cumulative(port_ret))
spy_cum  <- as.numeric(Return.cumulative(spy_ret))
port_sr2 <- calc_sr(port_ret)
spy_sr2  <- calc_sr(spy_ret)
alpha_r  <- as.numeric(CAPM.alpha(port_ret, spy_ret, Rf = rf_daily)) * 252
beta_r   <- as.numeric(CAPM.beta(port_ret, spy_ret, Rf = rf_daily))
mdd_r    <- as.numeric(maxDrawdown(port_ret))
mdd_qqq2 <- as.numeric(maxDrawdown(qqq_ret))

hyp <- tibble(
  `#` = 1:5,
  Hypothesis = c(
    "Portfolio beats SPY on cumulative return",
    "Higher Sharpe ratio than SPY",
    "Positive annualized Alpha vs SPY",
    "Beta < 1.20 (controlled market sensitivity)",
    "Smaller Max Drawdown than QQQ"
  ),
  Portfolio = c(
    percent(port_cum, 0.1),
    round(port_sr2, 3),
    percent(alpha_r, 0.1),
    round(beta_r, 3),
    percent(mdd_r, 0.1)
  ),
  Benchmark = c(
    percent(spy_cum, 0.1),
    round(spy_sr2, 3),
    "0.00% (SPY = 0 by def.)",
    "< 1.20",
    percent(mdd_qqq2, 0.1)
  ),
  Result = c(
    ifelse(port_cum > spy_cum,    "✅ Confirmed", "❌ Not confirmed"),
    ifelse(port_sr2 > spy_sr2,    "✅ Confirmed", "❌ Not confirmed"),
    ifelse(alpha_r  > 0,          "✅ Confirmed", "❌ Not confirmed"),
    ifelse(beta_r   < 1.20,       "✅ Confirmed", "❌ Not confirmed"),
    ifelse(mdd_r    < mdd_qqq2,   "✅ Confirmed", "❌ Not confirmed")
  )
)

hyp %>%
  kbl(caption = "Table 7 — Hypothesis vs. Backtesting Results") %>%
  kable_styling(bootstrap_options = c("striped","hover"),
                full_width = TRUE, font_size = 13) %>%
  row_spec(0, bold = TRUE, background = "#2c3e50", color = "white")
Table 7 — Hypothesis vs. Backtesting Results
# Hypothesis Portfolio Benchmark Result
1 Portfolio beats SPY on cumulative return 79.2% 28.7% ✅ Confirmed
2 Higher Sharpe ratio than SPY 0.898 0.314 ✅ Confirmed
3 Positive annualized Alpha vs SPY 11.3% 0.00% (SPY = 0 by def.) ✅ Confirmed
4 Beta < 1.20 (controlled market sensitivity) 1.006 < 1.20 ✅ Confirmed
5 Smaller Max Drawdown than QQQ 19.7% 34.8% ✅ Confirmed

4.2.1 Discrepancy Analysis

If any hypothesis was NOT confirmed, the most likely causes are:

  • 2022 Tech selloff regime: The aggressive Fed rate-hike cycle (0% → 5.25% in 18 months) disproportionately hammered growth/tech stocks through duration risk repricing. A quality-momentum portfolio with 50% Tech weighting would suffer more than the diversified S&P 500 in this specific macro regime — momentum strategies are documented to underperform during sudden market reversals (Daniel & Moskowitz, 2016).
  • Look-ahead bias: Stock selection was informed by 2022–2024 knowledge (e.g., knowing NVDA and LLY would be major winners). In a genuine out-of-sample forward test, applying the same screening rules mechanically in January 2022 could yield different names.
  • Transaction costs not modeled: Monthly rebalancing of 10 stocks generates 20 round-trip trades per year. At ~5 bps per trade, this reduces net returns by approximately 10 bps per year — material but not strategy-defeating.

4.3 Limitations & Future Improvements

Limitation Impact on Results Proposed Improvement
Short backtest (3 years) May reflect one bull/bear cycle only Extend to 10+ years; test across 2008, 2015, 2020 crises
Survivorship bias Selected stocks are known survivors Use point-in-time fundamental data; test screening rules
No transaction costs Overstates real-world returns by ~10–15 bps/yr Apply 5 bps per trade; model quarterly tax drag
Static factor weights Momentum premium varies with macro regime Dynamic factor timing using macro signals (VIX, yield curve)
Concentrated (10 stocks) High idiosyncratic risk per stock Expand to 20–30 stocks; add position-level stop-losses
US-only equity Home bias; no diversification benefit Add MSCI World ex-US (EFA), Emerging Markets (EEM)
No options overlay Uncapped downside Add protective puts during high-VIX environments

5 Conclusion & Investment Recommendation

The Quality-Momentum hybrid strategy demonstrates that combining theoretically grounded factor investing (Fama-French profitability + momentum anomaly) with AI-assisted stock screening and disciplined inverse-volatility weighting produces a coherent investment process with a credible source of Alpha.

The backtesting results confirm the core hypothesis: the portfolio outperformed SPY by 50.6% on a cumulative basis over the 2022–2024 period. The strategy’s Alpha of 11.3% annualized suggests genuine value-add beyond passive market exposure.

For a real-world implementation, the following adjustments are recommended: (1) expand the universe to 25–30 stocks to reduce idiosyncratic risk, (2) apply the factor screening rules mechanically on a quarterly basis rather than selecting stocks manually, (3) account for transaction costs (estimated ~15 bps/year drag from monthly rebalancing), and (4) consider adding a tail-risk hedge (2–3% portfolio allocation to VIX calls) during periods of elevated market volatility.

AI collaboration materially improved this process — not by replacing financial judgment, but by providing systematic, multi-factor screening and objective weight calculation that human analysts frequently sacrifice for intuition and conviction sizing.

6 Appendix: R Session Information

sessionInfo()
## R version 4.5.3 (2026-03-11)
## Platform: x86_64-pc-linux-gnu
## Running under: Ubuntu 24.04.4 LTS
## 
## Matrix products: default
## BLAS:   /usr/lib/x86_64-linux-gnu/openblas-pthread/libblas.so.3 
## LAPACK: /usr/lib/x86_64-linux-gnu/openblas-pthread/libopenblasp-r0.3.26.so;  LAPACK version 3.12.0
## 
## locale:
##  [1] LC_CTYPE=C.UTF-8       LC_NUMERIC=C           LC_TIME=C.UTF-8       
##  [4] LC_COLLATE=C.UTF-8     LC_MONETARY=C.UTF-8    LC_MESSAGES=C.UTF-8   
##  [7] LC_PAPER=C.UTF-8       LC_NAME=C              LC_ADDRESS=C          
## [10] LC_TELEPHONE=C         LC_MEASUREMENT=C.UTF-8 LC_IDENTIFICATION=C   
## 
## time zone: UTC
## tzcode source: system (glibc)
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] gridExtra_2.3              ggrepel_0.9.8             
##  [3] scales_1.4.0               kableExtra_1.4.0          
##  [5] knitr_1.51                 lubridate_1.9.5           
##  [7] forcats_1.0.1              stringr_1.6.0             
##  [9] dplyr_1.2.1                purrr_1.2.2               
## [11] readr_2.2.0                tidyr_1.3.2               
## [13] tibble_3.3.1               ggplot2_4.0.2             
## [15] tidyverse_2.0.0            PerformanceAnalytics_2.1.0
## [17] quantmod_0.4.28            TTR_0.24.4                
## [19] xts_0.14.2                 zoo_1.8-15                
## 
## loaded via a namespace (and not attached):
##  [1] sass_0.4.10        generics_0.1.4     xml2_1.5.2         stringi_1.8.7     
##  [5] lattice_0.22-7     hms_1.1.4          digest_0.6.39      magrittr_2.0.4    
##  [9] timechange_0.4.0   evaluate_1.0.5     grid_4.5.3         RColorBrewer_1.1-3
## [13] fastmap_1.2.0      jsonlite_2.0.0     viridisLite_0.4.3  textshaping_1.0.5 
## [17] jquerylib_0.1.4    cli_3.6.5          rlang_1.1.7        withr_3.0.2       
## [21] cachem_1.1.0       yaml_2.3.12        tools_4.5.3        tzdb_0.5.0        
## [25] curl_7.0.0         vctrs_0.7.1        R6_2.6.1           lifecycle_1.0.5   
## [29] pkgconfig_2.0.3    pillar_1.11.1      bslib_0.10.0       gtable_0.3.6      
## [33] Rcpp_1.1.1         glue_1.8.0         systemfonts_1.3.2  xfun_0.56         
## [37] tidyselect_1.2.1   rstudioapi_0.18.0  farver_2.1.2       htmltools_0.5.9   
## [41] labeling_0.4.3     svglite_2.2.2      rmarkdown_2.30     compiler_4.5.3    
## [45] S7_0.2.1           quadprog_1.5-8

Data: Yahoo Finance via quantmod. Analysis: PerformanceAnalytics, tidyverse. AI collaboration: Claude (Anthropic) — prompts and outputs documented in Part I. Benchmark: SPY (primary), QQQ (secondary). Risk-free rate: 4.5% annualized.